The Bartz v Anthropic fairness hearing on 14 May 2026 saw little resistance to the headline number. But the class definition reaches the registered US works problem. It does not reach the global corpus problem.

The Hearing

On 14 May 2026 the Northern District of California held the final fairness hearing in Bartz et al v Anthropic PBC, Case No. 3:24-cv-05417, the USD 1.5 billion proposed class settlement covering eligible registered works Anthropic allegedly downloaded from pirate libraries. Judge Araceli Martínez-Olguín took the matter under submission. She inherited the docket from Judge William Alsup, who retired at the end of 2025, after the June 2025 summary judgment that held training on lawfully acquired books to be fair use while leaving the pirated central-library theory exposed for trial.¹

By the date of the hearing 447,576 of approximately 482,460 eligible works had been claimed, a rate of 92.77 percent. Class counsel from Lieff Cabraser characterised the opt-out and objection rates as minimal. The judge’s questioning concentrated on attorneys’ fees and the settlement’s cost structure rather than the substantive deal terms.²

The Registration Trap

The most consequential objection came from author George Tombs, whose works were excluded because they lacked a US copyright registration. The class definition restricts membership to works that satisfy the settlement’s eligibility criteria, including qualifying US registration. The predictable consequence is that many foreign-authored books, and many unregistered US works, fall outside the settlement entirely.

The exclusion is statutory. Section 411(a) requires registration as a precondition for bringing a US infringement action in respect of a United States work. Section 412 restricts statutory damages and attorneys’ fees where registration was not timely made, and that remedy restriction applies to foreign and domestic works alike.³ Neither provision extinguishes the underlying copyright, which subsists automatically in any work qualifying under Article 5(2) of the Berne Convention.⁴ Foreign authors whose works lack qualifying US registration, and other unregistered rightsholders, retain whatever underlying copyright they have but sit outside the USD 1.5 billion settlement structure.

The Comparative Position

The German position has begun to be clarified by the OLG Hamburg judgment in Kneschke v LAION, 5 U 104/24, 10 December 2025.⁵ The court held that section 44b UrhG, which implements Article 4 of Directive (EU) 2019/790, can permit text-and-data mining for AI-training datasets where no valid machine-readable opt-out has been declared. It also held that LAION, as a non-commercial research organisation, could rely separately on section 60d UrhG.⁶ The judgment matters for foreign authors with respect to Anthropic because Anthropic is a commercial enterprise. Section 60d is unlikely to assist it, and section 44b is available only where rightsholders have failed to opt out in a machine-readable form. If the relevant copying occurred before machine-readable opt-out mechanisms were standardised or widely implemented, the German exposure will turn on how courts treat that timing problem.

In the United Kingdom, the Government’s March 2026 copyright and AI report confirmed that a broad text-and-data-mining exception with opt-out is no longer its preferred way forward.⁷ The section 29A research exception remains narrow. Where the relevant copying occurs within the UK, unauthorised commercial training may be actionable as primary infringement under section 16 of the Copyright, Designs and Patents Act 1988, subject to proof of copying, territorial nexus and any applicable exception.⁸ Getty Images v Stability AI is the principal English authority to date, but it left the training-stage question only partly answered because the primary copyright claim was narrowed by jurisdictional and evidential issues. UK collective management organisations and author groups are obvious potential claim-coordination vehicles, but the procedural route remains unsettled.

In France, the infringement baseline rests on article L122-4 of the Code de la propriété intellectuelle. The text-and-data-mining exceptions sit in articles L122-5 and L122-5-3, following implementation of the CDSM Directive by Ordonnance n° 2021-1518.⁹ The opt-out architecture is materially similar to Germany’s. France is an obvious forum for coordinated rights-holder action, but the procedural route remains to be seen.

The Strategic Implication

The unresolved question is whether the US per-work figure of approximately USD 3,000 sets a reference point for parallel European proceedings. There is no doctrinal reason it should. Damages in European jurisdictions are typically calculated by reference to a notional licence fee or to an account of profits, not to settlement values reached in unrelated US class actions. The reference may still operate informally. Anthropic’s commercial incentive will be to characterise the US settlement as substantial compensation already provided. The counter is that excluded European authors and unregistered rightsholders received nothing under that settlement and their underlying rights remain unimpaired.

If approved, Bartz will close the US registered-works class. It will not close the international copyright map. European proceedings may determine whether the global cost of the allegedly pirated training corpus remains a US-class-settlement number or becomes a materially larger cross-border exposure.

Footnotes

¹ Bartz et al v Anthropic PBC, Case No. 3:24-cv-05417 (ND Cal); Order on Motion for Summary Judgment, 23 June 2025 (Alsup, J.).

² Authors Alliance, “Bartz v. Anthropic Fairness Hearing: Observations and Takeaways”, 14 May 2026; Publishing Perspectives, “Anthropic Settlement Appears to Cruise Through Its Final Fairness Hearing”, 15 May 2026.

³ 17 U.S.C. sections 411(a), 412, 504(c).

⁴ Berne Convention for the Protection of Literary and Artistic Works, Article 5(2).

⁵ OLG Hamburg, Kneschke v LAION, Case No 5 U 104/24, 10 December 2025.

⁶ Directive (EU) 2019/790, Article 4; Urheberrechtsgesetz, sections 44b and 60d.

⁷ UK Government, Report and Impact Assessment on Copyright and Artificial Intelligence, March 2026.

⁸ Copyright, Designs and Patents Act 1988, sections 16 and 29A.

⁹ Code de la propriété intellectuelle, articles L122-4, L122-5 and L122-5-3; Ordonnance n° 2021-1518.

Introduction

On 18 April 2026, Sullivan & Cromwell LLP wrote to Chief Judge Martin Glenn of the United States Bankruptcy Court for the Southern District of New York after Boies Schiller Flexner identified AI-generated errors in a Chapter 15 filing in the Prince Global Holdings proceedings. The reported errors included inaccurate citations, misstatements or misquotations of bankruptcy law and other drafting defects that should have been caught before filing.¹ The incident matters not because elite firms are uniquely vulnerable, but because it shows that AI risk has moved from the margins of legal practice into the core workflow of sophisticated firms.

The professional liability question is becoming unavoidable: when does reliance on AI constitute negligence, and when might failure to use AI become negligent? Those questions appear to point in opposite directions. In fact, they are converging. The emerging standard is not anti-AI and it is not pro-AI. It is a standard of competent, supervised and verified use.

The Scale of the Problem

The Sullivan & Cromwell incident is not isolated. It is the most prominent recent example of a wider failure pattern. Damien Charlotin’s AI Hallucination Cases database tracks legal decisions in which generative AI has produced hallucinated content, or where AI use has been addressed by a court or tribunal in more than a passing reference. The database is important, but its methodology matters: it does not purport to capture every false citation in every filing. It captures the court-facing subset that has reached judicial or tribunal attention.²

As at 15 May 2026, Charlotin’s tracker identified 1,450 matters worldwide, including 1,003 in the United States.³ The precise number changes quickly. The direction of travel does not. What began as a few notorious filing failures has become a recurring operational risk in litigation practice.

The origin point in public consciousness remains Mata v Avianca. In June 2023, Judge P. Kevin Castel of the Southern District of New York imposed a $5,000 Rule 11 sanction after lawyers submitted an affirmation containing six non-existent judicial opinions generated by ChatGPT.⁴ At the time, much of the profession treated Mata as a cautionary novelty. By 2026, that reading is no longer sustainable.

The sanctions environment has hardened. In Couvrette v Wisnovsky, the District of Oregon found that summary judgment briefing contained 15 non-existent cases and eight fabricated quotations. The financial outcome was not a single fine, but a combination of a $15,500 sanctions order and later fee and cost orders that together exceeded $110,000.⁵ The distinction matters: the case should not be described simply as a ‘$110,000 fine’. Its real significance is broader. Courts are increasingly willing to make the cost of verification failure fall on the lawyers and parties responsible for putting false material before the court.

Court-level requirements are also multiplying. As at May 2026, Legal AI Governance identified 113 active orders and rules binding attorney filings.⁶ The requirements vary. Some require disclosure of AI use. Some require certification that AI-assisted work has been checked. Others simply restate that Rule 11, candour and professional responsibility duties apply irrespective of the tool used. The common message is that ‘AI did it’ is not a defence.

The Regulatory Response

The American Bar Association responded with Formal Opinion 512 on 29 July 2024, its first comprehensive ethics opinion on generative AI in legal practice. The opinion addresses competence, confidentiality, communication with clients, billing, candour to the tribunal and supervisory responsibilities under Model Rules 1.1, 1.4, 1.5, 1.6, 3.3, 5.1 and 5.3.⁷ The central proposition is simple: a lawyer may use generative AI, but the lawyer remains responsible for the work product.

Formal Opinion 512 does not require lawyers to understand the model architecture of every system they use. It does require them to understand the capabilities and limitations of the tools sufficiently to use them competently. It also requires lawyers to protect confidential information, communicate with clients where AI use materially affects the representation, verify AI-generated legal analysis and citations before submission, supervise junior lawyers and non-lawyers using AI and avoid billing clients for time spent correcting avoidable AI errors.

California has moved towards a more prescriptive model. In 2026, the State Bar of California sought public comment on proposed amendments to six Rules of Professional Conduct addressing AI: competence, communication, confidentiality, candour to the tribunal, managerial responsibility and supervision of non-lawyer assistants.⁸ If adopted, the proposals would make verification explicit. A proposed comment to the competence rule states that, when using technology including AI, a lawyer must independently review, verify and exercise professional judgment over the output. A proposed comment to the candour rule specifically requires verification of the accuracy and existence of cited authorities before submission to a tribunal.

That is not yet the same as an enacted rule. The California material should therefore be described as a proposed amendment, not as binding black-letter law. Its significance lies in the direction of regulatory movement: duties that were previously implicit in competence and candour rules are being translated into AI-specific drafting.

Colorado is useful for a different reason. It has both a disciplinary example and a rule-based development. In People v Crabill, a Colorado lawyer received a suspension of one year and one day, with 90 days actively served, after using ChatGPT-generated case law without reading or verifying it, failing to alert the court to sham cases and then falsely attributing the errors to a legal intern.⁹ Separately, the Colorado Supreme Court approved Rule Change 2026(02), which adds commentary making clear that technology, including AI, does not diminish a lawyer’s professional responsibilities and that a lawyer who uses technology in delivering legal services may be subject to discipline for a resulting rule violation.¹⁰

These regulatory materials share a common architecture. They treat AI as a tool, not as an independent professional actor. They impose verification duties. They reject any presumption that AI-generated legal output is reliable. They locate the obligation within existing competence, candour, confidentiality and supervision frameworks rather than creating a wholly new category of professional regulation.

The English Position

The leading English authority is Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank, heard together and decided by the Divisional Court in June 2025.¹¹ The judgment is often described as an AI case, but that shorthand needs care. The court was dealing with the actual or suspected use of generative AI by lawyers to produce written legal arguments or witness statements that were not checked, resulting in false information being put before the court.

In the Ayinde matter, the underlying problem was the inclusion of five fake cases in judicial review materials. The provenance of the false authorities was contested; the barrister denied using AI. Ritchie J nevertheless made wasted costs orders of £2,000 each against Ms Forey and Haringey Law Centre and required referral to the Bar Standards Board and the Solicitors Regulation Authority.¹² The Divisional Court later held that the threshold for initiating contempt proceedings was met in relation to Ms Forey, but decided not to initiate contempt proceedings or refer the case to the Law Officers.¹³

The practical guidance from the judgment is more important than the disputed factual mechanics. The court stated that freely available generative AI tools such as ChatGPT are not capable of conducting reliable legal research. They may cite sources that do not exist, quote passages that do not appear in genuine sources and produce confident assertions that are simply untrue. Lawyers who use AI for legal research, or rely on others who have done so, have a professional duty to check the output against authoritative sources before using it in advice or before a court.

The Solicitors Regulation Authority has not adopted an ABA-style AI ethics opinion. Its approach remains principles-based. On 22 April 2026, however, the SRA opened its consultation, ‘Strengthening our continuing competence approach’, running until 15 July 2026.¹⁴ The consultation proposes stronger requirements for solicitors to record how they identify and address learning and development needs and to participate in annual ethics discussions. It is not an AI-specific rulebook, but it sits alongside the SRA’s broader AI risk materials and compliance guidance on AI and technology.¹⁵

The Law Society’s guidance on generative AI likewise has persuasive rather than binding force. Its September 2025 update expressly added references to Ayinde and Al-Haroun, and it now stresses verification against reliable and authoritative sources, supervision and risk management.¹⁶ The English position is therefore less prescriptive than the emerging American model, but not less serious. The courts have made clear that false authorities in court documents may trigger wasted costs orders, regulatory referral and, in an appropriate case, contempt proceedings.

The Emerging Dual Standard

Professional liability in the age of generative AI is developing along two vectors at the same time. The first is negligent reliance. A lawyer who submits AI-generated legal material without proper verification may breach duties of competence, candour, supervision and care to the client. That proposition is now strongly supported by the cases, ethics opinions and regulatory materials.

The second vector is less developed but potentially more disruptive: the possibility that failure to use AI may, in some circumstances, fall below the standard of care. This proposition should not be overstated. No court has yet held a lawyer negligent simply for failing to use AI. Nor is there a general duty to use every available technology. The better formulation is narrower: where a particular AI tool has become a reasonably standard, reliable and proportionate means of improving the relevant task, a professional who fails to consider or deploy it may struggle to justify the omission if the client suffers avoidable loss.

That argument is consistent with orthodox professional negligence principles. The standard is not perfection. It is the standard of a reasonably competent professional in the relevant circumstances. In English law, Bolam and Bolitho provide useful analogies, particularly the idea that a practice accepted by a responsible body of professionals must also withstand logical scrutiny. In the solicitor context, the same essential question appears in a different form: what would a reasonably competent practitioner have done, judged by the standards of the profession at the time?

Anurag Bana’s SSRN paper on artificial intelligence, legal professional negligence and AI-covered indemnity risk articulates the point clearly: as AI becomes prevalent in legal practice, liability may arise both from using AI incorrectly and from failing to use it where its use would have been reasonably expected.¹⁷ Recent professional negligence commentary takes a similar position: AI is not generally mandatory, but the real question is whether a reasonably competent professional would have used it in the client’s interests in the particular circumstances.¹⁸

The point is easiest to see in verification. A lawyer who asks a public chatbot to produce case law and files the result without checking it is plainly exposed. But consider the inverse case: a firm has access to a reliable citation-checking, document-comparison or disclosure-analysis tool; the tool is widely used for the task; the cost of using it is proportionate; and the error that later causes loss is precisely the kind of error the tool would probably have caught. In that scenario, failure to use AI is not negligence because AI exists. It is negligence, if at all, because the professional failed to use an available and reasonably standard quality-control method.

The practical tension is acute. A lawyer who uses AI without verification may be negligent. A lawyer who refuses to use AI where it has become an ordinary part of competent practice may also become exposed. The safe ground is not abstention. It is disciplined adoption: use AI where it adds value, understand its limits, verify its output, supervise its use and preserve human professional judgment as the final decision-making layer.

Indemnity and Insurance Implications

Professional indemnity insurers now face the dual standard as both a coverage and pricing problem. The negligent reliance vector produces familiar claims in new clothing: failed applications, adverse costs orders, lost procedural opportunities, confidentiality breaches and client losses caused by unverified work product. The failure-to-use vector is more novel: claims alleging that a lawyer failed to use an available technology that would have prevented the loss.

Coverage questions remain unresolved. United States commentary on lawyers’ professional liability policies notes that many policies do not expressly exclude AI use, but coverage may depend on whether the conduct falls within the policy definition of professional services and whether exclusions for intentional acts, fraud, fee disputes or technology failures are engaged.¹⁹ Some professional liability insurers have also begun experimenting with AI-specific exclusions or endorsements.²⁰

The hardest cases will sit between negligence and recklessness. An isolated failure to check an AI-assisted draft may be characterised as negligence. Repeated submission of fabricated authorities after warning signs have been raised, or a deliberate refusal to verify citations known to have come from a generative tool, may be characterised very differently. Coverage will turn on policy wording, governing law and the factual findings in the underlying claim.

The NAIC Model Bulletin on the use of AI systems by insurers, adopted in December 2023, is relevant but only indirectly.²¹ It addresses insurer use of AI in insurance operations, including governance, risk management and compliance with insurance law. It does not solve the professional indemnity question of how insurers should underwrite or respond to claims arising from insured lawyers’ use, misuse or non-use of AI.

For underwriters, the risk is now two-sided. Traditionally, technology risk in professional indemnity was framed as a risk of using defective systems. AI introduces the additional possibility of technology abstention risk: the allegation that competent practice required the use of a tool and the insured failed to use it. That does not mean underwriters should require blanket AI adoption. It means proposal forms, renewal questions and risk engineering will need to move beyond asking whether a firm uses AI and start asking how AI is governed, supervised, verified and documented.

Strategic Outlook

The standard of care is moving. Expected work product quality will rise as reliable AI-assisted methods become normalised. Firms that ban AI entirely face one liability vector. Firms that permit uncontrolled AI use face another. Both approaches are inferior to governed use.

The first strategic imperative is verification. Every firm should have a clear rule that AI-generated legal authorities, quotations, factual propositions and analytical conclusions must be checked against authoritative sources before submission to a court, delivery to a client or use in advice. The rule should apply not only to partners and associates, but also to trainees, paralegals, knowledge teams, external consultants and anyone else contributing to legal work product.

The second imperative is supervision. AI use should be treated like any other delegated work stream. A partner is not excused because the first draft came from an AI tool rather than a junior lawyer. A recent US sanction against a managing partner for a junior lawyer’s AI-related citation error illustrates the same point in operational terms.²² The supervision question is the same: who checked it, against what source, using what process and where is the evidence that the check occurred?

The third imperative is training. The profession has spent too much time debating whether AI should be used and too little time teaching practitioners how to use it responsibly. Competence training needs to cover prompt design, tool selection, confidentiality, privilege, hallucination risk, citation verification, document comparison, disclosure workflows, billing treatment and escalation protocols. The SRA’s continuing competence consultation points in this direction even if it does not prescribe AI-specific rules.

The fourth imperative is insurance engagement. Firms should not wait for a claim before asking how their professional indemnity policy treats AI-assisted legal work. They should understand whether AI use is within the scope of covered professional services, whether any AI-specific exclusions apply, whether sanctions and fee-shifting orders are covered and what notification obligations arise when an AI-related error is discovered.

The dual standard is likely to sharpen over the next three to five years. Courts will continue to sanction negligent reliance. At some point, a failure-to-use allegation is likely to be tested in a professional negligence claim, most probably where a client can show that a readily available verification, disclosure or drafting tool would have caught the error that caused the loss. The claim may or may not succeed. Its arrival should surprise no one.

The profession cannot resolve this tension by choosing one vector over the other. It can resolve it only by occupying the disciplined middle ground: deploying AI tools competently, verifying their output rigorously, supervising their use systematically and maintaining human judgment as the irreducible core of legal practice. Firms that master that discipline will reduce cost, improve quality and protect themselves. Firms that do not will be exposed in both directions: for unverified AI reliance and, in time, potentially for failing to adopt standard AI-assisted checks.

Notes

1. Reuters, ’Sullivan & Cromwell law firm apologizes for AI hallucinations in court filing’, 21 April 2026; Legal Cheek, ’Sullivan & Cromwell apologises after AI hallucinations appear in court document’, 22 April 2026.

2. Damien Charlotin, AI Hallucination Cases Database, methodology note: database tracks legal decisions where the use of AI, whether established or merely alleged, is addressed in more than a passing reference by a court or tribunal; it does not track the wider universe of all false citations or uses of AI in court filings.

3. Damien Charlotin, AI Hallucination Cases Database, last updated 15 May 2026, identifying 1,450 cases worldwide and 1,003 USA matters. The database is live and should be rechecked immediately before publication.

4. Mata v Avianca, Inc., No. 22-cv-1461, 678 F Supp 3d 443 (SDNY 2023), sanctions order dated 22 June 2023.

5. Couvrette v Wisnovsky, No. 1:21-cv-00157-CL (D Or), Opinion and Order on sanctions, 12 December 2025, 2025 WL 4109655; Opinion and Order on fee/cost allocation, 23 March 2026, ECF No. 225; subsequent merits order, 30 March 2026, ECF No. 227.

6. Legal AI Governance, Federal and State Court Orders on AI tracker, identifying 113 active orders and rules binding attorney filings.

7. American Bar Association, Formal Opinion 512, Generative Artificial Intelligence Tools, 29 July 2024.

8. State Bar of California, Proposed Amendments to the Rules of Professional Conduct Related to Artificial Intelligence, public comment material, 2026; public comment deadline 4 May 2026.

9. People v Zachariah C. Crabill, 23PDJ067, Colorado Office of the Presiding Disciplinary Judge, 22 November 2023.

10. Colorado Supreme Court, Rule Change 2026(02), approved 8 January 2026.

11. Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank QPSC [2025] EWHC 1383 (Admin), Divisional Court, 6 June 2025.

12. R (Ayinde) v London Borough of Haringey [2025] EWHC 1040 (Admin), Ritchie J; summarised in the Divisional Court judgment at [2025] EWHC 1383 (Admin).

13. Ayinde and Al-Haroun [2025] EWHC 1383 (Admin), discussion of contempt threshold and decision not to initiate contempt proceedings.

14. Solicitors Regulation Authority, Strengthening our continuing competence approach, consultation opened 22 April 2026 and closing 15 July 2026.

15. Solicitors Regulation Authority, Risk Outlook report on the use of artificial intelligence in the legal services sector, updated 23 April 2026; SRA compliance tips on AI and technology, updated 9 February 2026.

16. The Law Society of England and Wales, Generative AI: the essentials, 1 October 2025, Updates: September 2025.

17. Anurag Bana, ’Artificial Intelligence, Legal Professional Negligence and the Rise of AI-Covered Indemnity Risk’ (SSRN, abstract dated 2025; PDF posted 9 April 2026).

18. Cripps, ’When is it negligent for a professional to use or ignore AI?’, 7 May 2026.

19. Reuters Legal News, ’From innovation to exposure: artificial intelligence risks for legal professionals’, 14 July 2025.

20. Reuters Legal News / Westlaw Today, ’Insuring against productive laziness: attorney use of artificial intelligence’, 22 December 2025.

21. National Association of Insurance Commissioners, Model Bulletin: Use of Artificial Intelligence Systems by Insurers, adopted 4 December 2023.

22. Reuters Legal News, ‘US judge says senior lawyers must pay for mistakes by subordinates using AI tools’, 1 May 2026.

Connecticut’s General Assembly passed SB 5 on 1 May 2026 by 131–17 in the House and 32–4 in the Senate. Governor Lamont’s office has signalled he will sign. If signed, the Connecticut Artificial Intelligence Responsibility and Transparency Act will become one of the most significant state AI packages to clear a US legislature in this cycle. It arrives in a landscape where the first such package, Colorado SB 24-205, has its enforcement temporarily suspended in federal court.

The Connecticut framework

SB 5 targets three domains. AI companions must disclose that the user is interacting with AI at the start of an interaction and at hourly intervals during continuous use.¹ Large generative AI providers above a one-million-user threshold face synthetic-content provenance obligations, with the C2PA technical standard identified as one relevant methodology.² Automated employment decision tools used as a substantial factor in hiring, promotion, discipline or discharge carry compliance obligations for developers and notice obligations for deployers.³

The effective dates are staggered. Core provisions begin on 1 October 2026. AI companion obligations begin on 1 January 2027. Public summaries diverge on the precise commencement of the synthetic-content and employment deployer obligations. The enrolled bill text controls. Freshfields places the provenance obligation on 1 October 2026 and the employment deployer obligation on 1 October 2027.⁴ Davis Wright Tremaine describes a broader watermarking obligation for synthetic digital content beginning 1 October 2027.⁵ DLA Piper describes SB 5 as a set of linked but distinct AI bills rather than a single governance statute.⁶ For in-scope providers, the legal effect may converge. The compliance date is what matters.

The Colorado litigation, the DOJ intervention and the enforcement pause

Colorado SB 24-205, the Colorado Artificial Intelligence Act, has a statutory effective date of 30 June 2026 after the delay enacted by SB 25B-004.⁷ The litigation posture has now overtaken the statutory timetable.

xAI commenced federal proceedings on 9 April 2026 in X.AI LLC v Weiser, No. 1:26-cv-01515 (D. Colo.), pleading First Amendment, Commerce Clause, due process and equal protection challenges across six counts.⁸ On 24 April 2026, the US Department of Justice lodged a statement of interest and intervened on Equal Protection grounds, focused on the bias-audit and algorithmic-discrimination architecture of the statute.⁹ On 27 April 2026, the District of Colorado granted a joint motion temporarily suspending enforcement of the Colorado AI Act pending further legislative or rulemaking developments and the outcome of xAI’s preliminary-injunction motion.¹⁰

That order alters the analytical frame for every state AI bill that follows the Colorado model. Connecticut is the immediate comparator. The dormant Commerce Clause and First Amendment lines that xAI is running against Colorado could be adapted to Connecticut’s provenance regime and AI companion disclosure regime, although the statutory architecture is different. The DOJ Equal Protection theory, however, is narrower and depends on the bias-audit architecture of the Colorado statute. Connecticut does not import that architecture in the same form.

The practitioner clock and Article 50

The EU AI Act, Regulation 2024/1689, Article 50(2), requires providers of AI systems generating synthetic audio, image, video or text content to ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. Article 50(4) imposes disclosure obligations on deployers producing content constituting a “deep fake”. The Article 50 obligations apply from 2 August 2026 under Article 113.¹¹

Connecticut’s provenance obligation, on the Freshfields reading, takes effect on 1 October 2026, two months after Article 50. Providers building C2PA-compatible provenance for EU compliance should be well placed to meet the technical component of Connecticut’s provenance regime if their user base brings them within scope, subject to the final enrolled text and implementing guidance. The thresholds are asymmetric. The EU obligations bind providers and deployers regardless of user size, with limited research and security exceptions. The Connecticut threshold is one million monthly active users. A smaller generative AI provider that escapes Connecticut will still be caught by Article 50.

The technical question of how to mark synthetic content is convergent across the EU, Connecticut and the C2PA-aligned platform stack. The legal exposure question is fragmented. The Colorado enforcement pause makes clear that state AI regulation is unstable before the second state’s first effective date arrives. Practitioners should treat state-law compliance calendars as operationally provisional, while treating provenance architecture, particularly C2PA-style implementation, as a likely direction of travel.

_____

¹ Connecticut SB 5, AI companion notice provisions; see DLA Piper, Unpacking SB5: Connecticut’s new AI law, May 2026.

² Connecticut SB 5, generative AI provenance provisions; see Freshfields, Connecticut Poised to Enact One of the Nation’s Most Comprehensive AI Laws; Transparency Coalition AI, TCAI Bill Guide: SB 5.

³ Connecticut SB 5, AEDT provisions; see Freshfields; Shipman & Goodwin, Connecticut’s AI Responsibility and Transparency Act: Key Impacts on the Workplace.

⁴ Freshfields, Connecticut Poised to Enact One of the Nation’s Most Comprehensive AI Laws.

⁵ Davis Wright Tremaine, Connecticut Adopts AI Transparency, Safety, and Consumer Protection Law.

⁶ DLA Piper, Unpacking SB5: Connecticut’s new AI law, May 2026: SB 5 described as “not a broad governance statute” but “a set of separate AI bills linked together”.

⁷ Colorado SB 24-205, Colorado Artificial Intelligence Act, effective date delay enacted by SB 25B-004, shifting commencement from 1 February 2026 to 30 June 2026.

⁸ X.AI LLC v Weiser, No. 1:26-cv-01515 (D. Colo. filed Apr. 9, 2026).

⁹ DOJ press release, Justice Department Intervenes in xAI Lawsuit Challenging Colorado’s Algorithmic Discrimination Statute, 24 April 2026.

¹⁰ Order, X.AI LLC v Weiser, No. 1:26-cv-01515 (D. Colo. Apr. 27, 2026), per Norton Rose Fulbright, xAI Sues, DOJ Intervenes, Enforcement of Colorado AI Act Suspended.

¹¹ EU Regulation 2024/1689 (the AI Act), Articles 50 and 113.

In November 2025, Amazon sued Perplexity AI in the Northern District of California over Comet, Perplexity’s AI browser agent. The case raises a deceptively simple question: can a user authorise an AI agent to access a password-protected marketplace on their behalf where the platform owner has expressly prohibited automated access? On 9 March 2026, Judge Maxine Chesney answered that question at the preliminary injunction stage in Amazon’s favour, holding that Amazon had shown strong evidence of unauthorised access. The Ninth Circuit’s subsequent administrative stay means the point is not yet settled. But the litigation has already framed the central legal issue for agentic commerce: the difference between user authority and platform authorisation.^5,6,17

The Amazon v. Perplexity litigation matters because it is beginning to articulate the boundary between user authority and platform control. Globally, the commercial architecture is shifting from traditional e-commerce to agentic commerce. This transition is marked by the delegation of consumer decisions to autonomous AI systems.^1,2 This shift moves beyond passive assistance to a state where AI agents sense their environment, plan multi-step workflows and execute transactions with little to no human supervision.^1,2 McKinsey research suggests that by 2030, agentic commerce could orchestrate between three trillion and five trillion dollars in global revenue, with the US retail sector alone accounting for up to one trillion.^3,4 However, this rapid integration of autonomous agents into commerce has outpaced existing legal structures, creating a “regulatory grey zone” across the United States, the United Kingdom and the European Union.³ The Amazon litigation now provides one of the first major attempts to map the boundaries of that zone.

The Conceptual Architecture of Agentic Commerce

To understand the legal challenges posed by agentic commerce, one must first distinguish these systems from earlier forms of automation. While earlier systems followed predefined, rule-based processes, agentic AI uses Large Language Models (LLMs) and reasoning frameworks to pursue high-level objectives over time.^2,7 These systems possess four core capabilities: autonomy from continuous supervision; goal-orientation; multi-step reasoning across disparate data sources; and the ability to act across multiple services or platforms.^2,7

In a commercial context, agentic systems operate at three levels of increasing sophistication. At the first level, agents interact directly with merchant platforms, mimicking human browsing to find and purchase goods. This pattern is termed “agent-to-site” transactions.¹ At the second level, systems transact with other agents, negotiating terms without human intermediaries.^1,8 At the third level, intermediary systems coordinate complex interactions across a global network of platforms, creating what may be called “multi-agent ecosystems.”¹

This shift moves the consumer from “using tools” to “delegating outcomes.”^2,9 Legally, this creates a “third actor problem.” The traditional bilateral relationship between consumer and merchant is complicated by an autonomous intermediary that may lack legal personality but can bind its principal to contracts, often without the principal knowing the specific terms.^3,10

US Litigation and the Precedent of Platform Authorisation

In the United States, the legal position on agentic commerce is being shaped primarily through the Computer Fraud and Abuse Act of 1986 (CFAA), a statute originally drafted to combat hacking that is now being tested against autonomous shopping agents.^11,12,13

The Test Case: Amazon.com Services LLC v. Perplexity AI, Inc.

In November 2025, Amazon filed suit against Perplexity AI in the Northern District of California, alleging that the developer’s “Comet” browser agent was configured to access password-protected Amazon customer accounts to browse products and complete purchases.^5,6,16 According to Amazon’s complaint, the agent deliberately concealed its automated nature by spoofing its User-Agent string to appear as a standard Google Chrome session, thereby evading Amazon’s bot detection systems.^5,14,15,16 Amazon contended that it issued at least five warnings and deployed technical barriers, which Perplexity bypassed within 24 hours.^5,16

Judge Maxine Chesney’s preliminary injunction rests on several key legal findings. First, the court ruled that user permission is not legally equivalent to platform authorisation. While a user may consent to an AI agent using their credentials, the platform owner retains the right to refuse that agent access to its infrastructure.^6,18 Second, the court found that by bypassing technical blocks and disguising its identity, Perplexity likely violated the CFAA’s prohibition on accessing a protected computer “without authorization.”^6,19,20 Third, the injunction rested partly on evidence that uncontrolled AI agents interfere with Amazon’s advertising systems, which require accurate detection of automated traffic to maintain billing integrity for advertisers.¹⁴

The Ninth Circuit Court of Appeals granted an emergency administrative stay of the injunction on March 16, 2026, allowing Perplexity to continue operations while the merits of the appeal are considered.^24,25,26 Perplexity’s defence argues that the CFAA does not prohibit accessing public websites and that the “access” in question is performed by the human consumer, not the company.²⁵

Evolution of the CFAA and Parallel Scraper Litigation

Amazon v. Perplexity represents the next step in a line of scraper cases. In hiQ Labs, Inc. v. LinkedIn Corp., the Ninth Circuit established that scraping publicly available data (where no login is required) does not constitute unauthorised access under the CFAA.^12,20,27 The Amazon case shifts the focus to “gated” environments. The court cited Facebook, Inc. v. Power Ventures, Inc., which established that once a platform owner explicitly withdraws permission via a cease-and-desist letter, any further access constitutes a CFAA violation even if account holders provide their credentials.^6,11,20 The distinction is not simply scraping versus non-scraping; it is public web access versus access to account-protected or technically restricted environments after permission has been withdrawn.

A further parallel is Ryanair v. Booking.com. In July 2024, a Delaware jury found that Booking.com violated the CFAA with an “intent to defraud” by screen scraping the airline’s website to resell tickets with unauthorised fees.^28,29 Ryanair’s win is consistent with a line of cases in which courts have been more receptive to platform-control arguments where access is technically restricted, commercial or contrary to express withdrawal of permission.^28,29

Privacy and Consumer Protection Claims in the US

Litigation in the United States is also extending into privacy and background-check statutes. In Ambriz v. Google LLC, plaintiffs allege that Google’s AI-powered Cloud Contact Center functions as a “third-party listener” that analyses call data without proper consent.^11,12 A separate class action filed in January 2026 targets Eightfold AI, alleging that its AI-powered hiring tools violate the Fair Credit Reporting Act (FCRA) and California’s Investigative Consumer Reporting Agencies Act (ICRAA).³⁰ The complaint argues that by assembling individualised reports from diverse internet sources to assess job suitability, the AI service becomes a “consumer reporting agency” subject to strict disclosure and accuracy requirements.³⁰

US courts have also begun sanctioning lawyers for submitting AI-generated filings that include fabricated case citations, as seen in Mata v. Avianca, Inc..^31,32,33 These rulings confirm that AI is a tool and that human professionals bear ultimate responsibility for the accuracy of information submitted under their name.^31,33

The UK Approach: Sectoral Regulation and Business Responsibility

The United Kingdom has avoided the EU’s path of horizontal statutory AI regulation, opting instead for a “pro-innovation,” sectoral approach.^34,35,36,37 Oversight is distributed among existing regulators: the Competition and Markets Authority (CMA), the Information Commissioner’s Office (ICO), and the Financial Conduct Authority (FCA).^9,36,38

CMA Guidance: Direct Accountability for AI Agents

On March 9, 2026, the CMA published detailed guidance on complying with consumer law when using AI agents.⁹ The regulator’s core position is that businesses cannot outsource their legal responsibilities to autonomous code; they are as accountable for the actions of their AI agents as they are for those of their human employees.^9,39

The CMA’s guidance can be reduced to three practical compliance duties. First, businesses must embed consumer protection principles into agent design, ensuring that agents respect statutory rights, provide accurate pricing, and avoid misleading omissions.^9,39,40 Second, businesses need appropriate human oversight and monitoring to catch hallucinations, inaccurate outputs and other failures that could mislead consumers or cause financial loss.^9,39,40 Third, if an agent is found to be non-compliant or making errors, the business must act immediately to refine prompts or workflows.^9,40 Failure to correct an identified breach promptly will result in accountability.

Under the Digital Markets, Competition and Consumer Act 2024 (DMCC Act), the CMA has the power to impose administrative fines of up to 10 per cent of a company’s global turnover for consumer law breaches, including misleading sales practices delivered by AI systems.^9,41

English Common Law and AI Liability

The UK Jurisdiction Taskforce (UKJT) launched a consultation in January 2026 on a draft Legal Statement on liability for AI harms under English private law.^42,43,44 The statement holds that under English law, AI does not have legal personality and therefore cannot be held responsible for physical or economic harm.^44,45,46 Liability must instead be attributed to legal persons through existing principles of negligence.^44,45,46

The UKJT analysis examines vicarious liability, asking whether an employer may be vicariously liable for an employee’s negligent use of AI, though generally not for the AI system itself.^45,46 It also considers professional standard of care, asking whether a professional may be liable for breach of duty by failing to perform due diligence on an AI system before using it for client work, and conversely, whether failure to use AI where a reasonable professional would do so might also constitute a breach.^45,46 The analysis further addresses factual causation. The “black box” nature of AI, where the internal reasoning behind a specific output is opaque, may require courts to approach causation differently through evidential assumptions or expert testimony.^46,47 Because the UKJT statement remains a draft consultation document, it should be treated as influential guidance on likely common-law analysis, not as a binding statement of law.

Hallucinations in UK Courts and Tribunals

By early 2026, the UK legal system had recorded at least 31 reported instances of AI hallucinations in litigation, including fictitious case citations and misrepresented legal norms.⁴⁸ In Taiwo v. Homelets of Bath Ltd (2025), the High Court issued a warning after a litigant submitted a skeleton argument referencing cases that did not exist.^49,50 The managing partner in Choksi v. IPS Law LLP (2025) submitted a witness statement found to contain fabricated authorities, leading to costs orders and potential referrals to regulators.^50,51 In Hassan v. ABC International Bank, a tribunal found 46 inaccurate citations (9 fictitious and 37 misrepresentations) and concluded the conduct was reckless and unreasonable, justifying a costs order.⁵⁰

The judiciary has updated its AI Guidance for Judicial Office Holders twice in 2025 to address these risks, emphasising that confidentiality must be maintained and that AI use must protect the integrity of the administration of justice.^48,52,53

The European Union: The Horizontal Statutory Model

In Brussels, the EU has enacted the world’s first horizontal law regulating Artificial Intelligence: the EU AI Act.^54,55,56,57 While the Act does not specifically name “agentic AI,” its definition of AI systems focuses on machine-based systems that operate with varying levels of autonomy and adaptiveness. This definition clearly encompasses autonomous agents.^58,59,60,61

The AI Act and High-Risk Agentic Systems

Under the EU AI Act’s tiered, risk-based approach, some agentic systems used in commerce may be high-risk, but only where they fall within the Act’s product-safety route or one of the Annex III use cases, such as creditworthiness, employment, access to essential services or other decisions affecting fundamental rights.^54,61,64 Ordinary retail discovery or checkout agents should not be described as high-risk merely because they are autonomous. One unresolved question is whether “human control” must be exercised in real-time at the moment of an autonomous purchase, or whether setting parameters in advance constitutes sufficient oversight.^3,54

Liability and the Revised Product Liability Directive

The Commission officially withdrew the proposed AI Liability Directive in 2025. The revised Product Liability Directive (PLD) now does part of the practical work that the AI liability package was expected to perform: it brings software, including AI systems, within the concept of a product and updates disclosure and evidential rules for technically complex claims.^{63,67,68,69,70} It should not, however, be described as a full replacement for the withdrawn AI Liability Directive.

Under the revised PLD, strict liability applies: manufacturers are liable for damages caused by product defects.^69,70 In cases of extreme technical complexity, courts may mandate disclosure of evidence and create a rebuttable presumption that the AI system was defective.⁶⁷ Importers from third countries are held to the same standards where their AI outputs are used within the Union.⁶⁷

Some commentators and payments-sector observers have suggested a future category of “supervised digital agents,” sitting between human users and autonomous machine actors.³ That idea is not yet a settled EU legislative category and should be presented as a possible future governance model rather than current law. The concept would formalise delegation credentials and tamper-evident behavioural logs for autonomous action systems.^3,67

GDPR and CJEU Clarifications on Automated Decision-Making

Under the General Data Protection Regulation (GDPR), AI agents processing personal data remain subject to Article 22, which provides individuals the right not to be subject to a decision based “solely on automated processing” that produces legal consequences or substantially affects their interests.^37,66,71,72

Recent CJEU case law has sharpened these rights. In the Schufa Case, the Court ruled that an automated creditworthiness score constitutes an automated “decision” if it plays a decisive role in a lender’s choice, even if a human nominally makes the final determination.^67,71 In the D&B Case, the Court held that meaningful information about the logic involved must be provided in an intelligible form, while trade-secret concerns may require procedural balancing rather than simple refusal.^71,73 Consumers also retain the right to challenge AI-driven outcomes and request human review whenever a decision may affect them.⁷⁴

These rulings indicate that as AI agents take on more commerce tasks (insurance risk assessment, dynamic pricing), they will face more demanding transparency and explainability requirements.^58,61,73

Systemic Effects and Strategic Implications of Agentic Commerce

Agentic commerce marks a turning point for the web, as it shifts from a human-readable medium to one that is machine-actionable.^4,75,76

The End of Permissionless Scraping and the Rise of AEO

The Amazon v. Perplexity ruling, if upheld on appeal, signals the beginning of the end for permissionless AI access to gated commercial platforms.⁷⁵ For years, parts of web scraping operated in a legal grey zone, but the court’s endorsement of platform control suggests that the practice of quietly extracting data and seeking forgiveness afterwards is drawing to a close.^22,77

For merchants, this requires a shift from Search Engine Optimization (SEO) to “Answer Engine Optimization” (AEO).^3,78 Brands must now make their product data, pricing, and return policies machine-readable so that AI agents can find, evaluate, and transact with them.^3,18,76 Cloudflare’s “Markdown for Agents” is an early technical response to this trend, allowing enabled sites to serve Markdown versions of HTML pages to AI agents through content negotiation.⁷⁶

The Race for Standardised Transactional Protocols

A competition is under way to define the communication standards for agentic commerce. Two primary models are emerging. Google’s Universal Commerce Protocol (UCP), launched in early 2026, establishes a common language between AI agents, merchant systems, and payment providers.^18,79,80,81 Google says UCP was co-developed with Shopify, Etsy, Wayfair, Target and Walmart and endorsed by more than 20 ecosystem participants, including Mastercard, Visa, Stripe, Adyen and The Home Depot.^80,81 It allows agents to identify themselves and transact through infrastructure that the retailer controls.^18,81

The second model is OpenAI and Stripe’s Agentic Commerce Protocol (ACP), which embeds the entire shopping journey within conversational interfaces like ChatGPT, with Instacart already offering “Instant Checkout” features via ACP.^79,80

Both protocols aim to reduce the friction that leads to abandoned carts while giving merchants a way to monetise bot traffic that might otherwise bypass traditional advertising surfaces.^79,82,83,84

Consumer Agency vs. Manipulative “Dark Patterns”

While agentic commerce offers genuine time savings and reduced decision fatigue for consumers, it also introduces new risks.² Consumer advocacy groups like Bureau Européen des Unions de Consommateurs (European Consumer Organisation) have warned that AI agents could steer users toward outcomes that benefit the developers or the retailers paying the highest commissions, rather than the consumer’s actual interest.^85,86,87

The CMA and European Commission are particularly concerned about “dark patterns” (interface designs that mislead consumers into decisions they would not otherwise make, such as false countdown timers or manufactured scarcity messaging).^39,88,89 In the agentic era, these patterns could become invisible to the human user, occurring as “agent-to-agent” interactions where one bot deceives another.⁸⁸

Algorithmic collusion also merits attention. Competition authorities in the UK and EU have made algorithmic pricing a priority.^90,91 The concern is that if multiple competitors use the same pricing algorithm or share commercially sensitive data through a common third-party provider, it could produce coordinated price increases without any explicit human communication. This phenomenon is termed “agentic collusion.”^9,90,91,92 In February 2026, the CMA launched an investigation into hotel chains suspected of sharing information through a common data services provider.⁹⁰ Polish competition authorities confirmed multiple ongoing investigations into collusion via algorithmic tools in the banking and pharmaceutical sectors in late 2025.⁹⁰ In the United States, Agri Stats litigation illustrates the same hub-and-spoke concern in a data-services setting, even though it is not itself an AI-agent case. In January 2026, Agri Stats agreed to resolve a meat-workers wage-suppression class action, and in May 2026 the DOJ and several states announced a proposed settlement of their meat-pricing case.^90,93

Conclusions and Strategic Implications

The transition to agentic commerce exhibits a widening gap between what the technology can do and what the law has settled on.³ The Amazon v. Perplexity litigation has, at least at the preliminary injunction stage, framed the central platform-access issue for agentic commerce: user-delegated authority is not necessarily the same thing as platform authorisation.^6,18 As agentic systems enter commercial deployment, the question before courts and regulators is shifting from “is it possible?” to “who is responsible when it fails?”^3,95,96

For stakeholders in the digital economy, four priorities warrant attention. First, AI developers should move toward “permissioned AI” architectures that secure both user credentials and explicit platform authorisation via official APIs. This “Dual Authorisation” model stands in contrast to the user-permission-only model that Perplexity attempted.^20,75,81

Second, organisations must implement regularly updated compliance records and detailed audit logs as required by the EU AI Act and recommended by the CMA.^{9,35,40,66,97} This includes real-time monitoring of agent behaviour to detect hallucinations, bias and potential collusive patterns before they generate serious liability.^9,39,97 Such “Governance by Design” embeds accountability into the systems themselves rather than relying on post-facto enforcement.

Third, businesses should be transparent when using AI agents, particularly where a consumer’s knowledge that they are dealing with an AI (rather than a human) would affect their decision to proceed.^9,39,40 “Transparent Consumer Disclosure” is not merely a compliance obligation; it also respects the autonomy of individuals making delegated purchasing decisions through agentic systems.

Fourth, companies should review and update technology contracts to address agentic transactions, defining the scope of authority granted to AI and setting out mechanisms for dispute resolution and indemnification.^1,98 Standard disclaimers written for passive software may no longer be legally adequate for autonomous systems that execute high-value contracts.⁹⁸ This “Contractual Risk Allocation” approach forces organisations to grapple explicitly with the liability surface created when machines bind their principals to commercial obligations.

As the global market for agentic commerce grows toward five trillion dollars by 2030, the legal frameworks of the United States, United Kingdom and European Union will continue to be tested.^3,4 The cases decided over the next two to three years will determine whether agentic commerce develops on a permissioned, regulated basis or whether the law trails the technology by a decade, as it has before.

References

¹ Digital commerce redefined: The growing impact of agentic commerce --- Linklaters / TechInsights. https://techinsights.linklaters.com/post/102lwgk/digital-commerce-redefined-the-growing-impact-of-agentic-commerce

² Agentic AI and consumers --- GOV.UK. https://www.gov.uk/government/publications/agentic-ai-and-consumers/agentic-ai-and-consumers

³ Agentic Commerce: When AI Buys on Your Behalf --- European Business Magazine. https://europeanbusinessmagazine.com/business/agentic-commerce-when-ai-buys-on-your-behalf-who-pays-whos-liable/

⁴ The agentic commerce opportunity --- McKinsey. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-agentic-commerce-opportunity-how-ai-agents-are-ushering-in-a-new-era-for-consumers-and-merchants

⁵ Judge blocks Perplexity’s AI bot from shopping on Amazon --- GeekWire. https://www.geekwire.com/2026/judge-blocks-perplexitys-ai-bot-from-shopping-on-amazon-in-early-test-of-agentic-commerce/

⁶ Court Finds AI Agent May Violate Federal Law by Accessing Amazon Accounts Without Authorization --- Cooley. https://www.cooley.com/news/insight/2026/2026-03-17-court-finds-ai-agent-may-violate-state-federal-law-by-accessing-amazon-accounts-without-authorization

⁷ Agentic AI in Financial Services: Regulatory and Legal Considerations --- Hogan Lovells. https://www.hoganlovells.com/en/publications/agentic-ai-in-financial-services-regulatory-and-legal-considerations

⁸ Amazon vs. Perplexity: A Lawsuit Over AI Agents that Shop --- Marketing AI Institute. https://www.marketingaiinstitute.com/blog/amazon-sues-perplexity

⁹ Free agent? New UK guidance on agentic AI for businesses --- Ashurst. https://www.ashurst.com/en/insights/free-agent-not-quite-new-uk-guidance-on-agentic-ai-for-businesses/

¹⁰ Agentic commerce: can the law of contract adapt? --- Gilbert + Tobin. https://www.gtlaw.com.au/insights/agentic-commerce-can-the-law-of-contract-adapt

¹¹ How 2 Tech Statutes Are Being Applied To Agentic AI --- Weil, Gotshal & Manges LLP. https://www.weil.com/-/media/files/pdfs/2026/february/law360\--how-2-tech-statutes-are-being-applied-to-agentic-ai.pdf

¹² AI Agents Are Raising New Questions of Fraud and Privacy Liability --- PYMNTS.com. https://www.pymnts.com/cpi-posts/ai-agents-are-raising-new-questions-of-fraud-and-privacy-liability/

¹³ Computer Fraud and Abuse Act (CFAA) --- NACDL. https://www.nacdl.org/Landing/ComputerFraudandAbuseAct

¹⁴ Amazon wins order blocking access by Perplexity’s AI shopping agent --- Reuters. https://www.reuters.com/legal/litigation/amazon-wins-order-blocking-access-perplexitys-ai-shopping-agent-2026-03-10/

¹⁵ Federal judge blocks Perplexity’s AI browser from making Amazon purchases --- CyberScoop. https://cyberscoop.com/amazon-perplexity-comet-browser-injunction/

¹⁶ Amazon vs. Perplexity AI: Court Blocks Comet Shopping Agent (2026) --- ALM Corp. https://almcorp.com/blog/amazon-vs-perplexity-ai-comet-shopping-agent-court-ruling/

¹⁷ Court Blocks Perplexity from Accessing Amazon in Agentic AI Lawsuit --- The Fashion Law. https://www.thefashionlaw.com/court-blocks-perplexity-from-accessing-amazon-systems-in-agentic-ai-lawsuit/

¹⁸ The Authorization Gap: What Amazon’s Perplexity Injunction Reveals --- Winning with Walmart. https://winningwithwalmart.com/the-authorization-gap-what-amazons-perplexity-injunction-reveals-about-the-commerce-layer-walmart-is-building-differently/

¹⁹ Amazon.com Services LLC v. Perplexity AI Inc., No. 3:25-cv-09514, Preliminary Injunction Order (N.D. Cal. Mar. 9, 2026).

²⁰ hiQ Labs, Inc. v. LinkedIn Corp., 31 F.4th 1180 (9th Cir. 2022); Facebook, Inc. v. Power Ventures, Inc., 844 F.3d 1058 (9th Cir. 2016).

²¹ Computer Fraud and Abuse Act Update: Supreme Court decides Van Buren v. U.S. https://hh-law.com/blogs/blog-intellectual-property-litigation-protection-and-prosecution-dtsa-ai-artificial-intelligence-lawyers/cfaa-van-buren-v-united-states/

²² Van Buren v. United States, 593 U.S. 374 (2021).

²³ Amazon.com Services LLC v. Perplexity AI Inc., Complaint, No. 3:25-cv-09514 (N.D. Cal. filed Nov. 2025).

²⁴ Appeals court temporarily pauses order blocking Perplexity’s AI shopping agent on Amazon --- CyberScoop. https://cyberscoop.com/perplexity-comet-ai-shopping-agent-amazon-lawsuit-ninth-circuit-stay/

²⁵ Court Temporarily Lifts Order Banning Perplexity From Amazon --- MediaPost. https://www.mediapost.com/publications/article/413555/court-temporarily-lifts-order-banning-perplexity-f.html

²⁶ Court temporarily allows Perplexity AI shopping ‘agents’ on Amazon --- Westlaw Today. https://today.westlaw.com/Document/Ia1b3afe0222811f1bf7dc89c10580585/View/FullText.html

²⁷ The Legal Landscape of Web Scraping --- Quinn Emanuel. https://www.quinnemanuel.com/the-firm/publications/the-legal-landscape-of-web-scraping/

²⁸ Ryanair Wins Legal Case Against Booking.com Over Screen Scraping --- Simple Flying. https://simpleflying.com/ryanair-wins-case-booking-screen-scraping-reselling-tickets/

²⁹ Ryanair wins Booking.com ‘screen scraper’ case --- Aviation Business News. https://www.aviationbusinessnews.com/industry-news/ryanair-wins-booking-com-screen-scraper-case/

³⁰ Groundbreaking Lawsuit Tests Whether AI Hiring Tools Trigger FCRA Compliance --- Ogletree. https://ogletree.com/insights-resources/blog-posts/groundbreaking-lawsuit-tests-whether-ai-hiring-tools-trigger-fcra-compliance/

³¹ Lawyer Fined for Using AI-Generated Legal Documents with Fake Citations --- Spellbook. https://www.spellbook.legal/learn/lawyer-fined-using-ai-legal-fake-citations

³² Mata v. Avianca, Inc., No. 22-cv-1461 (S.D.N.Y. 2023), sanctions order.

³³ Federal Court Rules Client’s AI-Generated Documents Not Privileged --- JD Supra. https://www.jdsupra.com/legalnews/federal-court-rules-client-s-ai-6386760/

³⁴ The Future of Legal AI in the UK: Trends to Watch in 2025 --- Qanooni. https://qanooni.ai/blog/the-future-of-legal-ai-in-the-uk-trends-to-watch-in-2025/

³⁵ AI regulation: a comparative overview of the UK, EU and US --- Stevens & Bolton. https://www.stevens-bolton.com/insights/102kd49/ai-regulation-a-comparative-overview-of-the-uk-eu-and-us/

³⁶ AI regulation in the UK: The role of the regulators --- Bird & Bird. https://www.twobirds.com/en/insights/2026/uk/ai-regulation-in-the-uk-the-role-of-the-regulators

³⁷ Agentic AI in the workplace --- Prettys Solicitors LLP. https://prettys.co.uk/articles/agentic-ai-in-the-workplace/

³⁸ AI and the FCA: our approach. https://www.fca.org.uk/firms/innovation/ai-approach

³⁹ Complying with consumer law when using AI agents --- GOV.UK. https://www.gov.uk/government/publications/complying-with-consumer-law-when-using-ai-agents/complying-with-consumer-law-when-using-ai-agents

⁴⁰ Agentic AI and consumer law: the CMA’s guidance for businesses --- Lewis Silkin LLP. https://www.lewissilkin.com/insights/2026/03/13/agentic-ai-and-consumer-law-the-cmas-guidance-for-businesses-102mmud

⁴¹ CMA launches first DMCCA enforcement cases --- RPC Legal. https://www.rpclegal.com/snapshots/consumer/winter-2025/cma-launches-first-dmcca-enforcement-cases-and-finalises-price-transparency-guidance/

⁴² UK Jurisdiction Taskforce launches consultation on liability for AI harms --- Henderson Chambers. https://www.hendersonchambers.co.uk/2026/01/22/uk-jurisdiction-taskforce-launches-consultation-on-liability-for-ai-harms/

⁴³ UK Jurisdiction Taskforce launches consultation on liability for AI harms --- Practical Law. https://uk.practicallaw.thomsonreuters.com/w-049-2237

⁴⁴ UKJT publishes consultation on its Legal Statement on Liability for AI Harms --- Bird & Bird. https://www.twobirds.com/en/insights/2026/uk/uk-jurisdiction-taskforce-publishes-consultation-on-its-legal-statement-on-liability-for-ai-harms

⁴⁵ UKJT consultation: Liability for AI harms under English private law --- CMS. https://cms.law/en/gbr/legal-updates/ukjt-consultation-liability-for-ai-harms-under-english-private-law

⁴⁶ UK Jurisdiction Taskforce consults on draft legal statement on liability for AI harms --- HSF Kramer. https://www.hsfkramer.com/notes/litigation/2026-01/uk-jurisdiction-taskforce-consults-on-draft-legal-statement-on-liability-for-ai-harms

⁴⁷ What are the litigation trends in 2026? --- Taylor Wessing. https://www.taylorwessing.com/en/insights-and-events/insights/2026/03/what-are-the-litigation-trends-in-2026

⁴⁸ Fake cases... Will hallucinations stop? --- Counsel Magazine. https://www.counselmagazine.co.uk/articles/fake-cases-will-hallucinations-stop-

⁴⁹ Taiwo v. Homelets of Bath Ltd \[2025\] EWHC 3173 --- Leathes Prior. https://www.leathesprior.co.uk/news/taiwo-v-homelets-of-bath-ltd-ors-2025-ewhc-3173-kb-03-december-2025

⁵⁰ AI Hallucination Cases Database --- Damien Charlotin. https://www.damiencharlotin.com/hallucinations/

⁵¹ The increasing legal liability of AI hallucinations --- VinciWorks. https://vinciworks.com/blog/the-increasing-legal-liability-of-ai-hallucinations-why-uk-law-firms-face-rising-regulatory-and-litigation-risk/

⁵² Artificial Intelligence (AI) --- Judicial Guidance (October 2025). https://www.judiciary.uk/guidance-and-resources/artificial-intelligence-ai-judicial-guidance-october-2025/

⁵³ Key developments in AI disputes in 2025 --- Taylor Wessing. https://www.taylorwessing.com/en/insights-and-events/insights/2026/01/key-developments-in-ai-disputes-in-2025

⁵⁴ EU AI Act: What It Means for Agentic Commerce --- Edgar, Dunn & Company. https://www.edgardunn.com/articles/the-new-rules-for-ai-inside-the-eus-bold-legislation

⁵⁵ EU AI Act 2026 Updates: Compliance Requirements and Business Risks --- Legal Nodes. https://www.legalnodes.com/article/eu-ai-act-2026-updates-compliance-requirements-and-business-risks

⁵⁶ What The EU AI Framework And UK’s Approach Mean For Your Agreements --- Docusign. https://www.docusign.com/en-gb/blog/what-the-eu-ai-framework-and-uks-approach-mean-for-your-agreements

⁵⁷ EU AI Act: first regulation on artificial intelligence --- European Parliament. https://www.europarl.europa.eu/topics/en/article/20230601STO93804/eu-ai-act-first-regulation-on-artificial-intelligence

⁵⁸ How to use agentic AI in line with the EU AI Act --- CX Network. https://www.cxnetwork.com/artificial-intelligence/articles/how-to-use-agentic-ai-in-line-with-the-eu-ai-act

⁵⁹ Agentic AI, Risk and Compliance Under the EU AI Act --- CMS.law (Sweden). https://cms.law/en/swe/legal-updates/agentic-ai-and-the-eu-ai-act2

⁶⁰ High-risk AI in the European Union --- DLA Piper Intelligence. https://intelligence.dlapiper.com/artificial-intelligence/?t=06-high-risk-uses&c=EU

⁶¹ Agentic AI, Risk and Compliance Under the EU AI Act --- CMS.law (Austria). https://cms.law/en/aut/legal-updates/agentic-ai-and-the-eu-ai-act2

⁶² Red Lines under the EU AI Act --- FPF. https://fpf.org/blog/red-lines-under-the-eu-ai-act-understanding-prohibited-ai-practices-and-their-interplay-with-the-gdpr-dsa/

⁶³ AI laws in 2026: the current EU--US landscape --- Blocshop. https://www.blocshop.io/blog/ai-laws-in-2026-the-current-euus-landscape-and-how-it-shapes-software-development

⁶⁴ Annex III: High-Risk AI Systems --- EU AI Act. https://artificialintelligenceact.eu/annex/3/

⁶⁵ Article 16: Obligations of Providers of High-Risk AI Systems --- EU AI Act. https://artificialintelligenceact.eu/article/16/

⁶⁶ EU AI Act vs GDPR: How They Work Together --- GuruSup. https://gurusup.com/blog/eu-ai-act-vs-gdpr

⁶⁷ Agentic law in the European Union: Governing autonomous AI agents --- Jurisconsul. https://www.jurisconsul.com/post/agentic-law-in-the-european-union-governing-autonomous-ai-agents

⁶⁸ 2026: the year AI grows up --- Taylor Wessing. https://www.taylorwessing.com/de/interface/2025/predictions-2026/2026-the-year-ai-grows-up

⁶⁹ The EU AI Act: What Businesses Need To Know --- Skadden. https://www.skadden.com/insights/publications/2024/06/quarterly-insights/the-eu-ai-act-what-businesses-need-to-know

⁷⁰ The Agentic AI Revolution: Managing Legal Risks --- Squire Patton Boggs. https://www.squirepattonboggs.com/insights/publications/the-agentic-ai-revolution-managing-legal-risks/

⁷¹ EU AI Act and GDPR: Tracing CJEU case law on automated processing --- Covington. https://www.cov.com/-/media/files/corporate/publications/2025/10/eu-ai-act-and-gdpr-tracing-cjeu-case-law-on-automated-processing-and-decision-making.pdf

⁷² Artificial intelligence --- UK Regulatory Outlook February 2026 --- Osborne Clarke. https://www.osborneclarke.com/insights/regulatory-outlook-february-2026-artificial-intelligence

⁷³ CJEU Clarifies GDPR Rights on Automated Decision-Making and Trade Secrets. https://www.insideprivacy.com/gdpr/cjeu-clarifies-gdpr-rights-on-automated-decision-making-and-trade-secrets/

⁷⁴ Consumers’ right to explanation under AI decision making --- BEUC. https://www.beuc.eu/sites/default/files/publications/BEUC-X-2026-003\_Consumers\_right\_to\_explanation\_under\_AI\_decision\_making.pdf

⁷⁵ Amazon Blocks Perplexity AI Scraping: Key Implications --- i10X. https://i10x.ai/news/amazon-perplexity-ai-scraping-lawsuit

⁷⁶ Cloudflare launches AI Audit and blocks unauthorized AI bots --- Cloudflare Blog. https://blog.cloudflare.com/ai-audit/

⁷⁷ Web scraping after hiQ v. LinkedIn: the legal landscape --- Perkins Coie. https://www.perkinscoie.com/en/news-insights/web-scraping-legal-landscape.html

⁷⁸ Amazon’s lawsuit against Perplexity rattles AI-driven search --- Hotel News Resource. https://www.hotelnewsresource.com/article139310.html

⁷⁹ Google’s Universal Commerce Protocol and the Race to Wire Agentic Shopping --- Opus Research. https://opusresearch.net/2026/01/15/googles-universal-commerce-protocol-and-the-race-to-wire-agentic-shopping/

⁸⁰ Agentic commerce AI tools, protocol for retailers, platforms --- Google Blog. https://blog.google/products/ads-commerce/agentic-commerce-ai-tools-protocol-retailers-platforms/

⁸¹ Walmart Partners with Google To Pioneer Agent-Led Commerce --- RetailWire. https://retailwire.com/discussion/walmart-google-agentic-commerce/

⁸² Agentic AI poses new challenges around online payments --- Pinsent Masons. https://www.pinsentmasons.com/out-law/news/agentic-ai-challenges-online-payments

⁸³ Should Amazon Be Banning AI Shopping Agents From its Platform? --- RetailWire. https://retailwire.com/discussion/should-amazon-banning-ai-shopping-agents/

⁸⁴ Amazon Injunction Could Change the Future of Agentic Commerce --- PYMNTS.com. https://www.pymnts.com/amazon/2026/amazon-injunction-could-change-the-future-of-agentic-commerce/

⁸⁵ Watchdog Issues Warning About Letting AI Run Your Life --- Futurism. https://futurism.com/artificial-intelligence/ai-agent-manipulation

⁸⁶ EU: BEUC published position paper on consumers’ right to explanation --- DataGuidance. https://www.dataguidance.com/news/eu-beuc-published-position-paper-consumers-right

⁸⁷ Consumers’ right to explanation under AI decision making --- BEUC. https://www.beuc.eu/position-paper/consumers-right-explanation-under-ai-decision-making

⁸⁸ AI shopping agents: How will UK Consumer Law apply? --- CMS. https://cms.law/en/gbr/legal-updates/ai-shopping-agents-how-will-uk-consumer-law-apply

⁸⁹ Complying With the DSA, AI Act, and GDPR --- Goodwin Law. https://www.goodwinlaw.com/en/insights/publications/2025/07/insights-practices-antc-three-laws-one-challenge

⁹⁰ When Bots Set Prices: CMA Highlights Real World Risks of Algorithmic Pricing --- Akin Gump. https://www.akingump.com/en/insights/alerts/when-bots-set-prices-cma-highlights-real-world-risks-of-algorithmic-pricing

⁹¹ AI and collusion: frontiers, opportunities and challenges --- CMA. https://competitionandmarkets.blog.gov.uk/2026/03/04/ai-and-collusion-frontiers-opportunities-and-challenges/

⁹² EU Regulations Are Not Ready for Multi-Agent AI Incidents --- TechPolicy.Press. https://www.techpolicy.press/eu-regulations-are-not-ready-for-multiagent-ai-incidents/

⁹³ US Justice Dept settles Agri Stats case --- Reuters. https://www.reuters.com/world/us-justice-dept-settles-agri-stats-case-2026-05-07/

⁹⁴ UKJT Public Consultation: Liability for AI Harms under the Private Law of England and Wales --- LawtechUK. https://lawtechuk.io/ukjt/public-consultation-liability-for-ai-harms-under-the-private-law-of-england-and-wales/

⁹⁵ 2026 AI Legal Forecast: From Innovation to Compliance --- Baker Donelson. https://www.bakerdonelson.com/2026-ai-legal-forecast-from-innovation-to-compliance

⁹⁶ Could agents be the next stumbling block for Europe’s AI rules? --- Euractiv. https://www.euractiv.com/news/could-agents-be-the-next-stumbling-block-for-europes-ai-rules/

⁹⁷ Driving compliance with EU’s AI Act through Agentic AI agents --- Consultancy.eu. https://www.consultancy.eu/news/12432/driving-compliance-with-eus-ai-act-through-agentic-ai-agents

⁹⁸ Agentic AI: The liability gap your contracts may not cover --- Clifford Chance. https://www.cliffordchance.com/insights/resources/blogs/talking-tech/en/articles/2026/02/agentic-ai-and-the-liability-gap-your-contracts-may-not-cover.html

Five publishers and Scott Turow sued Meta Platforms and Mark Zuckerberg on 5 May 2026 in the Southern District of New York for willful copyright infringement in the development of the Llama models¹. The complaint pleads that Zuckerberg and other Meta executives authorised and directed the torrenting of more than 267 TB of pirated material, including from LibGen, Anna’s Archive, Sci-Hub and other pirate sites. Three points are doing the analytical work in the complaint. Forum is the most important.

The Forum Is the Strategy

Three cases now frame the book-training branch of the AI copyright docket. Bartz v Anthropic² in the Northern District of California, where Judge Alsup held that Anthropic’s training use was fair use but refused to excuse the separate acquisition and retention of pirated works from LibGen and PiLiMi. The $1.5 billion settlement received preliminary approval on 25 September 2025 and remains subject to final approval. Kadrey v Meta³, also NDCal, where Judge Chhabria granted Meta summary judgment on the record before him on 25 June 2025 with the express warning that the ruling did not establish that Meta’s use of copyrighted works to train Llama was lawful as a general proposition. And now Elsevier v Meta in the Southern District of New York.

The publishers’ choice of venue is not coincidental. SDNY now has its own active AI copyright docket, including The New York Times v OpenAI before Judge Stein⁴. SDNY judges have not opined on the acquisition/use split that Judge Alsup applied in Bartz. Filing east of the Sierra Nevada gives the publishers a clean bench.

The Pleading Choices

Two pleading choices structure the case.

First, the complaint centres acquisition. Judge Alsup’s reasoning in Bartz turned on a clean conceptual division. Training on legally acquired copies was treated as transformative under Campbell v Acuff-Rose. The antecedent unlawful download of seven million books from LibGen and PiLiMi was independently actionable. The Elsevier complaint mirrors that division. It alleges over 267 TB of pirated material was torrented after Zuckerberg and other Meta executives authorised and directed the process, an executive-level decision in April 2023 to abandon publisher licensing negotiations and a knowing infringement theory under 17 U.S.C. § 504(c). Statutory damages on willful infringement can run to $150,000 per eligible work. If the allegations are proved, even at a fraction of that ceiling the exposure dwarfs the Anthropic settlement of $3,000 per work across roughly 500,000 books.

Second, the complaint names Zuckerberg personally. Naming Zuckerberg is not cosmetic. The complaint pleads direct infringement against him and, in the alternative, contributory infringement. The Rule 12 fight will therefore be whether the pleaded facts show knowledge, authorisation and material contribution, rather than merely apex status at Meta. The doctrinal frame remains Gershwin Publishing v Columbia Artists Management (2d Cir. 1971), Sony, Grokster and In re Aimster.

The Comparative Lens

UK practitioners are watching Getty Images v Stability AI for a different reason. The High Court’s 2025 judgment did not decide the central training-copying question under UK copyright law. It rejected important parts of Getty’s case and found only limited trade mark infringement. The comparison with Elsevier v Meta is therefore structural, not doctrinal: both cases turn on source material, provenance and what can be proved about training data. The US dispute is filtered through fair use. The UK dispute is not, but that does not mean UK claimants can ignore proof of copying, territoriality or subsistence.

EU regulatory attention has centred on the AI Act’s training-data transparency obligation under Article 53 of Regulation (EU) 2024/1689⁵, in force for general-purpose AI providers since 2 August 2025 with transitional rules for pre-existing models. Article 53 may produce public-facing and regulatory material relevant to provenance disputes, but it is not a substitute for US discovery. It is an evidential pressure point, not an evidence pipeline.

What to Watch in the Next Ninety Days

Three near-term decisions matter. First, the motion-to-dismiss schedule on the personal claims against Zuckerberg. Second, any consolidation or transfer motion under 28 U.S.C. § 1404 that could pull the case back to NDCal. Third, whether any party seeks coordination or consolidation of related SDNY AI copyright actions under Rule 42(a).

Anthropic showed why the acquisition record can drive settlement economics. The Bartz architecture has now travelled east. The economic question is whether SDNY will produce a Bartz-style holding without trial, or whether Meta will litigate the alleged willful-infringement record and the corporate-officer point to judgment.

Notes

¹ Elsevier Inc., Cengage Learning, Inc., Hachette Book Group, Inc., Macmillan Publishing Group, LLC, McGraw Hill LLC, Scott Turow and S.C.R.I.B.E., Inc. v Meta Platforms, Inc. and Mark Zuckerberg, Civil Action No. 26-cv-3689 (S.D.N.Y., complaint filed 5 May 2026).

² Bartz et al. v Anthropic PBC, No. 3:24-cv-05417 (N.D. Cal.) (Alsup, J.). Summary judgment opinion on fair use 23 June 2025. Settlement preliminary approval 25 September 2025; final approval pending (fairness hearing listed 14 May 2026).

³ Kadrey et al. v Meta Platforms, Inc., No. 3:23-cv-03417 (N.D. Cal.) (Chhabria, J.). Summary judgment for Meta on training fair use 25 June 2025.

⁴ The New York Times Co. v OpenAI, Inc. and Microsoft Corp., No. 1:23-cv-11195 (S.D.N.Y.) (Stein, J.).

⁵ Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 (the AI Act). Article 53 obligations on providers of general-purpose AI models entered into application on 2 August 2025, with transitional rules for models placed on the market before that date.

On 27 April 2026, CoinDesk reported that CFTC Chairman Michael Selig said the Commission is using, or building, artificial intelligence tools to assist with crypto registration applications.¹ The reported use case is administrative triage: flagging incomplete filings, identifying clearly deficient submissions and helping prioritise the staff queue. The reporting also indicates that AI is being applied to trading data and to surveillance for fraud, market manipulation and insider trading. Reporting around Selig’s 16 April 2026 appearance before the House Agriculture Committee linked the deployment of AI tools to reduced staffing and to the agency’s need to supervise expanding crypto and prediction-market responsibilities.²

Triage at the regulatory gateway

Two structural facts give the disclosure analytical weight. The CFTC is the primary federal regulator for crypto derivatives. Following the SEC/CFTC joint interpretation of 17 March 2026, the Commission has committed to administering the Commodity Exchange Act consistently with the SEC’s interpretation of when a non-security crypto asset may become, or cease to be subject to, an investment contract.³ AI is therefore operating close to the gateway of regulatory classification, registration and market access, at the moment when the boundary between securities-law treatment and commodity-law treatment of crypto assets is the most active question in US crypto regulation.

The 1946 framework

Section 706 of the Administrative Procedure Act (APA), enacted in 1946, requires reviewing courts to set aside agency action found to be arbitrary, capricious, an abuse of discretion, contrary to law, procedurally defective or, in some cases, unsupported by substantial evidence, on the basis of the whole record.⁴ Motor Vehicle Manufacturers Association of US Inc v State Farm Mutual Automobile Insurance Co, 463 US 29 (1983), held that the agency must articulate a satisfactory explanation for its action including a rational connection between the facts found and the choice made. The administrative record fixes the basis of judicial review.

When the agency’s first-pass review is automated, the question is what the record contains. A flag triggered by an AI tool is not, on its own, an explanation. A rejection rationalised as “clearly deficient” without a human articulation of which fact controlled cannot satisfy State Farm in the form courts have applied since 1983. The Supreme Court’s abandonment of Chevron deference in Loper Bright Enterprises v Raimondo, 144 S Ct 2244 (2024), reinforces the point: reviewing courts are less likely to accept agency characterisation by default where the legal basis for action is contested.

Comparative jurisdiction

The EU has addressed comparable public-law risk through classification. Annex III of the AI Act covers certain public-sector AI uses, including those used by public authorities to assess eligibility for essential public services and to assist in the administration of justice. Such systems are high-risk. Article 14 requires that high-risk systems be designed and developed to enable effective human oversight during use, with the burden falling on providers and deployers.⁵ Registration triage of the kind the CFTC is now performing is not necessarily mapped one-for-one onto Annex III, but the architectural framework for classifying public-sector AI is in place.

The UK has adopted the Algorithmic Transparency Recording Standard. Current GOV.UK guidance treats it as mandatory for central government departments and certain arm’s-length bodies within scope, where tools significantly influence decision-making with public effect or directly interact with the public, and recommends it more broadly across the public sector. The United States has federal AI governance guidance, but no equivalent statutory high-risk classification regime and no APA-specific transparency framework for automated regulatory triage. The disclosure of CFTC AI use was made in a press interview rather than a notice-and-comment proceeding, which is itself a procedural data point.

Strategic implication

For practitioners advising crypto applicants, the immediate question is whether to plead the issue. A denied or stalled application that has been triaged by AI presents a State Farm record problem. Until the Commission publishes its protocols, denial letters should be treated as potentially actionable on APA grounds, not merely as procedural setbacks. Litigation pressure will drive disclosure of model documentation, training data and human review thresholds. That pressure is the most likely path to a US analogue of public-sector AI oversight, arrived at through judicial review rather than legislation.

¹ CoinDesk, “CFTC’s AI Will Review U.S. Crypto Registration Applications, Chairman Tells CoinDesk” (27 April 2026).

² CoinDesk, “U.S. CFTC’s Selig Says AI Has Helped Make Up for Staffing Cuts at Key Crypto Watchdog” (16 April 2026); House Agriculture Committee hearing transcript, oral evidence of Chairman Michael Selig, 16 April 2026.

³ SEC Press Release 2026-30, “SEC Clarifies the Application of Federal Securities Laws to Crypto Assets” (17 March 2026).

⁴ 5 U.S.C. § 706.

⁵ Regulation (EU) 2024/1689 of 13 June 2024, Articles 6 and 14 and Annex III. High-risk obligations are currently scheduled to apply from 2 August 2026.

By late 2024, Waymo’s robotaxi fleet had logged more than 25 million fully autonomous public-road miles across the United States.[50] Yet a single accident in which an autonomous vehicle fails to predict or avoid a collision forces the legal system to confront its most consequential allocation problem: who bears responsibility when the dynamic driving task was performed not by a human but by software?[1] Autonomous vehicles do not create liability from nothing. They move the liability question from the driver’s conduct at the moment of impact to the design, deployment, monitoring and insurance of the system that performed the dynamic driving task. For over a century, legal frameworks governing road traffic anchored liability in human agency. The “driver” was the central locus of both control and responsibility.[1, 2] As the dynamic driving task (DDT) moves from human cognition to algorithms, the traditional negligence model is no longer sufficient on its own. It remains relevant, but it now operates alongside product liability, insurance-first compensation, safety-case regulation and data-based accident reconstruction.[2, 3, 4]

The emerging pattern is not the abolition of liability, but its redistribution. In Level 1 and Level 2 systems, liability remains substantially anchored in human supervision. At Levels 3 and 4, the legal analysis moves to the operational design domain, the handover or fallback architecture, the entity responsible for the automated driving system (ADS), the insurer-first compensation mechanism and the evidential record generated by the vehicle. The central question is no longer simply whether a driver was negligent, but whether the system was authorised, safely deployed, properly monitored, adequately updated and capable of proving what it did.[4, 5, 6]

Across the United States, the United Kingdom, Europe, Asia, and Australasia, regulators face the “pacing problem”: the inherent lag between technological innovation and legal frameworks that can regulate it.[7] Some jurisdictions have adopted “light-touch” guiding principles to foster innovation. Others have enacted proactive statutes that redefine the nature of vehicle operation and insurance.[7, 8, 9, 10] This analysis examines these diverging and converging frameworks, focusing on the mechanisms of liability, the role of data in accident reconstruction, and the evolution of motor insurance in an era of automated mobility.

The Hierarchical Framework of Automation and Legal Agency

Central to liability analysis is the taxonomy of driving automation established by the Society of Automotive Engineers (SAE). The SAE defines six levels ranging from no automation (Level 0) to full automation (Level 5).[11, 12, 13] SAE levels are not liability rules. They are the technical vocabulary through which statutes, insurers and courts decide whether the human, the ADS, the operator or the product supplier was responsible for the relevant part of the driving task.[5]

Classification of Driving Automation and Liability Thresholds

Most consumer vehicles currently available in ordinary use remain Level 1 or Level 2 driver-assistance systems, with the human driver responsible for monitoring and control.[11] The liability frontier lies in Level 3 and Level 4 systems, where the ADS may perform the dynamic driving task within an operational design domain (ODD). In these systems, the human may still be required to act as a “fallback-ready user” or “user-in-charge”.[14, 16, 17]

This transitional phase introduces a zone of legal ambiguity, particularly when the system requests the human to resume control. The determination of whether a human had a “reasonable time” to react to such a demand will be central to future litigation.[14, 18] Courts will have to define the temporal and cognitive boundaries of human transition readiness - a question with little direct precedent in conventional traffic law.

The United States: A Laboratory of Jurisdictional Patchworks

In the United States, the regulatory environment reflects a distinctive tension between federal safety authority and state-level statutory experimentation.[19, 20] To date, no comprehensive federal statute governs autonomous vehicle civil liability. The definition of a vehicle “operator” varies sharply across state lines.[20]

Federal Safety Authority and the Standing General Order

At federal level, the United States regulates autonomous vehicles (AVs) primarily through vehicle safety authority rather than a unified civil-liability statute. The National Highway Traffic Safety Administration’s (NHTSA) Standing General Order requires identified manufacturers and operators to report certain crashes involving ADS or Level 2 advanced driver-assistance system (ADAS) vehicles. The agency uses that data to identify safety concerns, investigate potential defects and, where necessary, require unsafe vehicles or equipment to be remedied.[11, 19, 20, 44]

Federal Motor Vehicle Safety Standards (FMVSS) were originally designed for vehicles with human controls. Manufacturers seeking to test vehicles without steering wheels or brake pedals must often request exemptions.[19, 20] Federal preemption (where federal law would override inconsistent state laws) remains unresolved. Current and earlier versions of the SELF DRIVE Act, most recently reintroduced by Representative Bob Latta in February 2026, have yet to clarify this jurisdictional boundary.[19]

State Statutory Innovation: California, Florida, and Arizona

As of late 2024, more than 35 states have enacted legislation or executive orders related to autonomous vehicles.[19, 20] These statutes represent the front line of legal evolution, with states adopting markedly different philosophies on liability and operator responsibility.

California has established one of the world’s most rigorous regulatory frameworks. Vehicle Code Sections 38750 through 38755 govern testing and deployment.[18, 21] Manufacturers must obtain specific permits and maintain $5 million in insurance or proof of self-insurance.[21, 22] California’s 2026 AV regulations, announced by the DMV on 28 April 2026, introduce a “Notice of AV Noncompliance” mechanism. Contemporary reporting indicates that, from 1 July 2026, law enforcement will be able to issue notices for driverless-vehicle traffic violations, triggering DMV investigation and potential permit consequences.[45, 49]

Florida adopted a contrasting approach intended to attract manufacturers.[19] Florida’s House Bill 311 (2019) explicitly designates the ADS itself as the “operator” when engaged. This decouples liability from human presence.[19] The law permits operation of vehicles without a licensed human occupant, provided the vehicle meets federal safety standards.[19] By legally designating the system as the operator, Florida created a statutory path for shifting liability from individuals to technology-deploying entities.[19]

Arizona’s framework is driven primarily by executive orders, particularly Executive Order 2018-04, which mandates that all automated driving systems comply with federal and state safety standards.[19] Arizona has positioned itself as a “test bed” for innovation. It established a designated oversight committee and requires companies to register their intended ODD.[19]

The Evolution of Tort and Product Liability in US Case Law

In the absence of a unified federal liability statute, US courts apply traditional tort principles and product liability theories to autonomous vehicle accidents.[4] A visible shift has occurred: from “driver error” as a cause of action to “design defect” or “failure to warn”.[6] In cases involving semi-autonomous systems like Tesla’s Autopilot, plaintiffs have successfully argued that manufacturers bear liability for systems that fail to monitor driver engagement adequately or that encourage foreseeable misuse. In Benavides v Tesla, Judge Beth Bloom denied Tesla’s post-trial motion after a jury award of $242.57 million, including $200 million in punitive damages, in a defective-design and product-liability case concerning Autopilot.[6, 46]

As vehicles reach Levels 4 and 5, claims are likely to be framed increasingly through product-liability concepts, treating the ADS not as a legal driver but as part of a product and service system whose design, warnings, updates and operational limits can be tested in litigation.[6]

Negligent entrustment and related owner-responsibility theories may also evolve where owners or operators fail to maintain software updates, sensors or calibration requirements.[6]

The United Kingdom: A Proactive Statutory Model

Westminster has positioned itself as a global leader in autonomous vehicle regulation through the Automated and Electric Vehicles Act 2018 (AEVA) and the subsequent Automated Vehicles Act 2024.[10, 24] Its approach is distinctly proactive, seeking to resolve “regulatory disconnection” before mass commercialisation of self-driving technology.[7]

The Automated and Electric Vehicles Act 2018: The “Insurer-First” Model

AEVA 2018 introduced a novel insurance and liability model for connected and autonomous vehicles (CAV).[7] It resolves the question of “who is liable” by establishing a “single insurance” policy covering both driver and vehicle.[7]

Under the AEVA 2018, if an accident is caused by an automated vehicle “driving itself,” the insurer is primarily liable to compensate any third party who suffers damage, including the “driver” who may be an occupant not actively driving.[7, 25] This first-instance insurer liability ensures swift compensation. The injured party need not identify or prove a specific software or sensor failure before recovering.[14]

Once the insurer has compensated the victim, the Act provides a statutory right of recourse. The insurer can recover costs from the party responsible for the accident, such as the vehicle manufacturer or a third-party software provider.[7, 14] This two-stage process protects consumers while holding technology providers accountable through insurance subrogation.[7]

The Automated Vehicles Act 2024 and the ASDE

Building on a four-year review by the Law Commission of England and Wales and the Scottish Law Commission, the Automated Vehicles Act 2024 established a framework for safe deployment of self-driving vehicles on British roads.[10, 24] The legislation introduces key legal roles and responsibilities that redefine the boundaries of liability.

A “bright line” distinction runs through the 2024 Act: driver support features on one side, self-driving systems on the other.[24] When a vehicle is authorised as “self-driving” and the ADS is engaged, the human “user-in-charge” is explicitly exempted from criminal responsibility for the dynamic driving task.[24] This marks a turning point in road traffic law. It removes the traditional burden of “care and control” from the human occupant and places responsibility for the way the authorised self-driving feature drives, for regulatory purposes, on the Authorised Self-Driving Entity (ASDE), while third-party compensation remains channelled primarily through the AEVA 2018 insurer-first model.[14, 24] However, the user-in-charge remains responsible for “non-driving” tasks such as ensuring the vehicle is insured and that passengers wear seatbelts.[14]

Both Law Commissions recommended that ASDEs disclose safety-relevant data to regulators and insurers.[14] This duty of transparency is intended to ensure that civil claims are decided fairly and that systemic risks are identified and mitigated before they cause further accidents.[14]

Continental Europe: Harmonising AI and Product Liability

Across the EU, autonomous vehicle liability is being reshaped by a broader initiative to update the legal framework for the digital age. Key drivers include modernisation of product liability law and regulation of artificial intelligence.[26, 27, 28]

The Revised Product Liability Directive (EU) 2024/2853

The most consequential development in European liability law is the revision of the Product Liability Directive (PLD), which came into force on 8 December 2024. Member States have until 9 December 2026 to transpose it; the old PLD continues for products placed on the market or put into service before then.[27] Once implemented, the new directive expressly brings software, including AI systems and related digital services, within the strict product-liability framework, whether embedded in a device or accessed via cloud services.[26, 27, 28]

Key innovations of the revised PLD include several changes that reshape liability across the product lifecycle. Software is now treated as a product and manufacturers of AI systems and digital components are liable for defects in software, including those arising from algorithmic errors or data quality issues.[6, 26] Lifecycle liability has expanded beyond the moment a product is “placed on the market.” Manufacturers who retain control through software updates or digital services remain liable for defects that arise after the product enters service.[28, 29] Providers of “connected digital services” that are essential for vehicle function (such as real-time navigation or mapping services) can be held jointly and severally liable alongside the vehicle manufacturer.[26, 29] Any entity that substantially modifies a product becomes liable as a “manufacturer” for the modified components.[26, 28, 29]

The directive also addresses information asymmetry by alleviating the burden of proof for victims.[27, 28] Given the limited explainability of AI, courts are now empowered to order disclosure of technical evidence.[28] If a manufacturer fails to comply with a disclosure order, the defectiveness of the product is presumed.[28] If technical complexity makes it excessively difficult for a claimant to prove a defect, the claimant only needs to show that it is likely the product was defective or that a causal link exists.[28]

Germany: The StVG Amendments and the Technical Supervisor

Germany, as Europe’s leading automotive manufacturer, implemented specific domestic laws to support Level 3 and Level 4 automation.[16, 30] The German Road Traffic Act (StVG) was amended in 2017 to allow conditionally automated driving and further amended in 2021 to create a framework for autonomous driving without a driver present.[16, 30]

A unique feature of the German framework is the “Technical Supervisor” (Technische Aufsicht).[12, 16] For Level 4 vehicles, while no driver is required on board, a natural person must monitor the vehicle externally. That person must be capable of deactivating it or authorising specific manoeuvres if the system encounters an error.[12, 16] The Technical Supervisor is not treated as the driver, but may be exposed under general tort principles if their monitoring or intervention falls below the required standard.[16]

Germany maintains a strict “holder” liability system, but with increased caps.[16] The maximum liability for the “vehicle holder” (usually the owner) in accidents involving automated systems has doubled to €10 million for personal injury and €2 million for property damage.[16] This ensures victims have access to substantial compensation through the owner’s mandatory insurance. The insurer retains a right of recourse against the manufacturer if a product defect is identified.[16]

The EU AI Act and Its Interplay with Liability

Many AI systems used in automated driving will be treated as high-risk under the EU AI Act (Regulation (EU) 2024/1689) where they are safety components of vehicles or vehicle systems subject to EU type-approval or conformity assessment.[6, 28] The AI Act is not itself a civil-liability statute, but its duties on risk management, data governance, transparency, logging, human oversight, robustness and cybersecurity may inform the applicable standard of care. Non-compliance may therefore become relevant evidence, although claimants will still need to establish the elements of the civil claim under national law.[6, 28]

Asia: State-Led Coordination and the Supervisory Model

Asia has emerged as a high-growth region for autonomous vehicles. China, Japan, and South Korea have each adopted state-led strategies that prioritise safety assurance and commercial deployment.[31, 32]

China: Shenzhen’s Milestone and National Taxonomy

China does not yet have a comprehensive national AV liability regime. General tort, product liability, transportation accident liability and insurance rules remain the baseline nationally.[32] The national standard for the classification of driving automation (GB/T 40429-2021) aligns closely with SAE levels and the “Safety Guideline for the Use of Autonomous Vehicles in Transportation Services” (2023) establishes the groundwork for commercialisation, but specific liability allocation remains a matter of local regulation and general law.[32]

Shenzhen’s 2022 “Regulations on the Administration of Intelligent and Connected Vehicles” provided the first concrete local rules for intelligent and connected vehicle (ICV) management.[3] The Shenzhen regulations clarify responsibility at different automation levels. For vehicles with a driver, the driver remains primarily responsible for traffic violations.[3] For driverless ICVs, the owner or user is responsible, but they have a clear right of recourse against the manufacturer if a system defect caused the accident.[3] This model shifts risk from vehicle damage and third-party liability to product and operational liability.[3]

Japan: Specified Automated Operation and the 2027 Roadmap

Japan’s policy roadmap targets Level 4 deployment on public roads nationwide by 2027, though this remains a stated ambition rather than a legislated deadline.[31] To support this ambition, Japan amended its Road Traffic Act in 2022 to define “Specified Automated Operation” (SAO).[34] SAO is legally separated from “driving,” allowing for driverless transportation services under strict permission systems.[34, 35]

Under Japanese law, the “SAO implementer” must obtain permission from the Prefectural Public Safety Commission and appoint an “SAO Supervisor”.[34] This supervisor is responsible for remote monitoring and must ensure that a person is sent to the scene of any accident to take necessary measures.[34] Japan’s approach maintains a hierarchical attribution framework where liability is distributed based on the negligence level of involved parties. However, increasing emphasis falls on the manufacturer’s responsibility for cybersecurity and system integrity.[2, 35]

South Korea: The Accident Investigation Committee

South Korea enacted the “Act on Promotion and Support of Commercialization of Autonomous Vehicles” (AVA) to create a supportive environment for the industry.[36, 37] Like the UK, South Korea amended its “Act on Guarantee of Automobile Accident Compensation” in 2020 to ensure that victims are first covered by the vehicle owner’s insurance.[35, 37]

A key feature of the South Korean system is the “Accident Investigation Committee” under the Ministry of Land, Infrastructure and Transport (MOLIT).[36] This committee is responsible for collecting and analysing data from the vehicle’s mandatory “autonomous driving information recording device” (DSSAD) to identify the technical cause of an accident.[36] This government-led investigation process is designed to provide swift relief to victims and provide an objective basis for the insurer’s subrogation claims against manufacturers.[36] In April 2026, Korea launched an Autonomous Vehicle Accident Liability Task Force to develop standards for accident responsibility and compensation procedures ahead of broader commercialisation.[48]

Australasia: Pragmatic Harmonisation and Safety Duties

Australia and New Zealand have approached autonomous vehicle liability through the lens of national harmonisation and adaptation of existing transport laws.[39, 47]

Australia: The National Automated Vehicle Safety Law (AVSL)

The Australian National Transport Commission (NTC) has led a programme to develop an “end-to-end” regulatory framework.[39, 47] In 2022, infrastructure and transport ministers agreed to develop a National Automated Vehicle Safety Law (AVSL). Current laws do not yet allow automated vehicles on public roads in the manner contemplated and the AVSL remains under development as part of a nationally consistent framework.[39, 47]

The primary subject of regulation under the proposed AVSL is the “Automated Driving System Entity” (ADSE), the corporation that assumes responsibility for the ADS.[39, 47] Australian ministers have reached consensus that the ADSE would be “legally in control” of the vehicle when the ADS is operating.[39, 47] The developing AVSL will enforce a “General Safety Duty” on ADSEs, requiring them to manage in-service safety risks “so far as is reasonably practicable”.[40] This approach moves away from a pure fault-based model toward a “safety assurance” model. The entity must prove it has followed best practices in development and deployment.[40]

New Zealand: The “Technology-Taker” Strategy

New Zealand has historically adopted a more observational stance, describing itself as a “taker of technology” that evaluates the success of other jurisdictions before committing to its own regulations.[13, 41] However, the Ministry of Transport has initiated a work programme to clarify where responsibility sits across the SAE levels.[13]

A primary concern in New Zealand is that current offence provisions are directed almost entirely toward human drivers.[13] The Ministry is exploring whether to adapt existing regulations or create a “bespoke set” for intelligent transport systems.[9] New Zealand is reviewing vehicle inspection settings, including the treatment of safety technologies, but its AV liability framework remains at work-programme rather than enacted comprehensive-regime stage.[41]

Cross-Cutting Determinants of Liability: Data and Insurance

Regardless of jurisdiction, the practical adjudication of autonomous vehicle accidents depends on two factors: the availability of high-fidelity data and the evolution of motor insurance products.

Data Storage: EDR vs. DSSAD

Regulators are mandating specialised data recorders to resolve the “black box” problem of AI decision-making.[42]

Event Data Recorders (EDR) and Data Storage Systems for Automated Driving (DSSAD) serve different evidentiary purposes. EDR provides the “how” of a crash: speed, braking, g-forces. DSSAD provides the “who”: was the system engaged, did it issue a transition demand, was the driver attentive.[42] Jurisdictions including California and South Korea have already mandated these systems as a precondition for deployment.[18, 42]

The Insurance Pivot: From Frequency to Severity

The insurance industry, led by global insurers such as Allianz, is recalibrating its models for the “new mobility era”.[43] Industry consensus holds that while accident frequency will decline substantially (potentially by 35 percent by 2040), the severity and cost of each claim will increase.[43]

This shift is driven by multiple factors. Technological complexity means that a minor bumper impact that once cost €500 to repair now involves replacement and recalibration of expensive light detection and ranging (LiDAR) and radar units, driving repair costs higher.[43] Product liability has shifted. Insurers are moving from insuring “human error” to insuring “system performance,” which requires closer collaboration with original equipment manufacturers (OEMs).[43] Cyber risk is emerging as a new underwriting category. New products are being designed to cover emerging threats such as software failures and cyberattacks that could lead to mass-collision events.[43]

Conclusion: The Path Toward Global Synthesis

The global review of autonomous vehicle liability reveals not the abolition of traditional doctrine, but its layering. Liability is being reallocated through insurance-first compensation, product liability, safety duties and data obligations. In place of the individual driver as the sole bearer of liability, a composite model is emerging where the manufacturer, the software developer and the fleet operator each bear responsibility calibrated to their role in the system.

The United States continues to rely on state-level statutory experimentation and judicial evolution. The United Kingdom and South Korea have provided a blueprint for “victim-first” insurance models that prioritise social stability over immediate fault-finding. The European Union has redefined the concept of a “product” itself to ensure that software is no longer a legal shield for manufacturers.

The successful deployment of autonomous vehicles will ultimately depend on “trust through accountability”.[43] For this trust to be realised, legal frameworks must ensure three outcomes: data access remains unrestricted for accident investigation, insurance models are sufficient to cover systemic failures and the “bright line” between human and machine control remains unambiguous.

As we approach 2030, the convergence of international standards will likely form a new “Lex Automatica”: a global body of law that balances the immense safety potential of automation with the enduring legal requirement for justice and compensation.

References

1 Navigating Liability in the Age of Autonomous Vehicles, https://www.wshblaw.com/experience-navigating-liability-in-the-age-of-autonomous-vehicles

2 Liability for Autonomous Vehicle Torts: Who Should Be Held Responsible? - MDPI, https://www.mdpi.com/2032-6653/16/12/665

3 Urban Transport of China Legislation of Intelligent Connected Vehicles Management and Innovative Practice in Shenzhen, https://www.chinautc.com/upload/accessorychinautc/20243/20243271415151704641.pdf

4 Comparing Tort Liability Frameworks in Autonomous Vehicle Accident Governance, https://www.researchgate.net/publication/399538965_Comparing_Tort_Liability_Frameworks_in_Autonomous_Vehicle_Accident_Governance

5 Comparing Tort Liability Frameworks in Autonomous Vehicle Accident Governance - MDPI, https://www.mdpi.com/2032-6653/17/1/32

6 Who’s Liable When AI Takes the Wheel? New Frontiers - Jones Day, https://www.jonesday.com/-/media/files/publications/2025/10/civil-liability-and-risk-mitigation-strategies-for-autonomous-vehicles/files/who-is-liable-when-ai-takes-the-wheel-white-paper/fileattachment/who-is-liable-when-ai-takes-the-wheel-white-paper.pdf?rev=c3f3cde57338496d9c5015c13e586480

7 Automated and Electric Vehicles Act 2018: An Evaluation in light of Proactive Law and Regulatory Disconnect, https://ejlt.org/index.php/ejlt/article/view/702/966

8 Regulation 2025, https://www.transport.govt.nz/assets/Uploads/Report/Reg-2025-Scenarios-and-Findings.pdf

9 Regulation 2025 - Emerging insights, https://www.transport.govt.nz/assets/Uploads/Report/Regulation-2025-Emerging-Insights.pdf

10 Automated and Electric Vehicles Act 2018 regulatory report July 2023 – December 2024 - GOV.UK, https://assets.publishing.service.gov.uk/media/67cf248ed38a67eb3afaa669/automated-and-electric-vehicles-act-2018-regulatory-report-july-2023_december-2024.pdf

11 Automated Vehicle Safety - NHTSA, https://www.nhtsa.gov/vehicle-safety/automated-vehicles-safety

12 Automated and Autonomous Driving. Legal Framework. - Mercedes-Benz Group, https://group.mercedes-benz.com/technology/autonomous-driving/driving/legal-framework.html

13 Automated Vehicles Work Programme - Ministry of Transport, https://www.transport.govt.nz/area-of-interest/technology-and-innovation/autonomous-vehicles-work-programme

14 Automated vehicles – Law Commissions drive forward regulatory..., https://www.dlapiper.com/insights/publications/2022/02/automated-vehicles-law-commissions

15 LIABILITY FOR DAMAGES CAUSED BY AUTONOMOUS DRIVING VEHICLES FROM THE EU LAW PERSPECTIVE, https://ojs.srce.hr/eclic/article/download/38091/18173/165820

16 Expert Guide: Autonomous Vehicles Law in Germany - CMS, https://cms.law/en/int/expert-guides/cms-expert-guide-to-autonomous-vehicles-avs/germany

17 Autonomous Vehicles - Ministry of Transport, https://www.transport.govt.nz/assets/Uploads/Ministry-of-Transport-AVs-background-paper-two-International-Developments.pdf

18 California Code, Vehicle Code - VEH § 38750 - Codes - FindLaw, https://codes.findlaw.com/ca/vehicle-code/veh-sect-38750/

19 Autonomous Vehicles | Self-Driving Vehicles Enacted Legislation, https://www.ncsl.org/transportation/autonomous-vehicles

20 The Current State of Self-Driving Car Regulations in the U.S. - Urban SDK, https://www.urbansdk.com/resources/the-current-state-of-self-driving-car-regulations-in-the-u-s

21 Autonomous vehicles law and regulation in California, United States - CMS, https://cms.law/en/int/expert-guides/cms-expert-guide-to-autonomous-vehicles-avs/california-united-united-united

22 Title 13, Division 1, Chapter 1 Article 3.7 – Testing of Autonomous Vehicles § 227.00. Purpose. (a) The regulations in this a - California DMV, https://www.dmv.ca.gov/portal/uploads/2020/06/Adopted-Regulatory-Text-2019.pdf

23 California’s Autonomous Vehicle Noncompliance Notices: A Defense Strategy Guide for Manufacturers - Bulldog Law, https://www.thebulldog.law/california-s-autonomous-vehicle-noncompliance-notices

24 Automated vehicles - Law Commission, https://lawcom.gov.uk/project/automated-vehicles/

25 Automated and Electric Vehicles Act 2018 - Legislation.gov.uk, https://www.legislation.gov.uk/ukpga/2018/18

26 New German Product Liability Regime: Broader Scope, Potentially Higher Exposure, https://www.gtlaw.com/en/insights/2026/1/new-german-product-liability-regime-broader-scope-potentially-higher-exposure

27 Liability for defective products - European Commission, https://single-market-economy.ec.europa.eu/single-market/goods/free-movement-sectors/liability-defective-products_en

28 Navigating product liability in high-security sectors: Addressing AI ..., https://www.whitecase.com/insight-alert/navigating-product-liability-high-security-sectors-addressing-ai-driven-risks-under

29 Germany: Update On Product Liability Law | A&O Shearman - JDSupra, https://www.jdsupra.com/legalnews/germany-update-on-product-liability-law-8120247/

30 Legal framework for automated and autonomous driving and teledriving in the EU and Germany - Taylor Wessing, https://www.taylorwessing.com/en/insights-and-events/insights/2026/02/legal-frameworks-for-autonomous-driving-and-teledriving

31 Regulations for Autonomous Vehicles: Where Do Countries Stand in 2024-2030? (Global Policy Trends) | PatentPC, https://patentpc.com/blog/regulations-for-autonomous-vehicles-where-do-countries-stand-in-2024-2030-global-policy-trends

32 Expert Guide: Autonomous Vehicles Law & Regulation in China, https://cms.law/en/int/expert-guides/cms-expert-guide-to-autonomous-vehicles-avs/china

33 Amendment to the Road Traffic Act for L4 automated driving National Police Agency of Japan - UNECE Wiki, https://wiki.unece.org/download/attachments/351698946/WP.1%20IWG-SUAT-12-02%20%28J%29%20Amendment%20to%20the%20Road%20Traffic%20Act%20for%20Lv4%20AD%20%28NPAofJP%29.pdf?api=v2

34 2025 Global Guide to Autonomous Vehicles, https://www.city-yuwa.com/wp/wp-content/uploads/2025/05/2025-Global-Guide-to-Autonomous-Vehicles-Japan-Chapter-.pdf

35 Recent trends in regulations on autonomous vehicles in Korea, https://www.ibanet.org/article/19FCDD11-A0B1-41F1-97AB-F32E144311F8

36 act on the promotion of and support for commercialization of autonomous vehicles - Statutes of the Republic of Korea, https://elaw.klri.re.kr/eng_service/lawViewContent.do?hseq=69828

37 COMPULSORY MOTOR VEHICLE LIABILITY SECURITY ACT, https://elaw.klri.re.kr/eng_service/lawView.do?hseq=70198&lang=ENG

38 2024–27 National Connected and Automated Vehicle (CAV) Action ..., https://www.infrastructure.gov.au/sites/default/files/documents/2024-27-national-connected-and-automated-vehicle-action-plan.pdf?

39 In-service safety for automated vehicles - National Transport Commission, https://www.ntc.gov.au/sites/default/files/assets/files/NTC-Decision-RIS-In-service-safety-for-AVs.pdf

40 National Transport Commission, General Safety Duty for Automated Vehicles - Decision Regulation Impact Statement, https://www.ntc.gov.au/transport-reform/automated-vehicle-program

41 Automated Vehicles Work Programme - Ministry of Transport New Zealand, https://www.transport.govt.nz/area-of-interest/technology-and-innovation/autonomous-vehicles-work-programme

42 Study on Standardization of Data Retrieval Tools for DSSAD based on SAE J1698 - The Korean Society of Automotive Engineers, http://journal.ksae.org/_common/do.php?a=full&b=22&bidx=4273&aidx=47352

43 Allianz Motor Day 2025: Hands Off – The Safety Promise of Autonomous Driving, https://www.allianz.com/content/dam/onemarketing/azcom/Allianz_com/press/document/motor-day-2025-report-hands-off-safety-promise-autonomous-driving.pdf

44 NHTSA Standing General Order on Crash Reporting, https://www.nhtsa.gov/laws-regulations/standing-general-order-crash-reporting

45 California DMV, New Autonomous Vehicle Regulations Strengthen Oversight and Enforcement (28 April 2026), https://www.dmv.ca.gov/portal/news-and-media/new-autonomous-vehicle-regulations-strengthen-oversight-and-enforcement-authorize-trucks-and-transit/

46 Benavides v Tesla Inc, Case No. 1:2021cv21940 (SD Fla), Post-Trial Order (Judge Beth Bloom), https://law.justia.com/cases/federal/district-courts/florida/flsdce/1%3A2021cv21940/593426/612/

47 Office of Future Transport Technology - Automated Vehicles, Australian Government, https://www.infrastructure.gov.au/infrastructure-transport-vehicles/transport-strategy-policy/office-future-transport-technology/automated-vehicles

48 Seoul Economic Daily, Korea Launches Task Force on Autonomous Vehicle Accident Liability (7 April 2026), https://en.sedaily.com/finance/2026/04/07/korea-launches-task-force-on-autonomous-vehicle-accident

49 Los Angeles Times, California can ticket robotaxis that violate traffic laws (1 May 2026), https://www.latimes.com/california/story/2026-05-01/california-can-ticket-robotaxis-that-violate-traffic-laws-heres-how

50 Waymo, New Swiss Re study: Waymo is safer than even the most advanced human-driven vehicles (December 2024), https://waymo.com/blog/2024/12/new-swiss-re-study-w

On 23 April 2026 Tether froze $344 million USD₮ across two TRON addresses in coordination with OFAC and US law enforcement.1 The following day OFAC updated Bank Markazi’s SDN List entry by adding two TRX digital currency addresses, with linkages to the IRGC-Qods Force and Hizballah.2 Treasury Secretary Scott Bessent framed the wider sanctions campaign as part of Economic Fury.3

From Citibank to TRON: the architectural shift

The $1.75 billion seizure underlying Bank Markazi v Peterson depended on assets held in a New York bank account, a statute, 22 U.S.C. section 8772, that identified the relevant enforcement proceeding and judicial enforcement under the Foreign Sovereign Immunities Act. The 2026 action depended on none of these. Tether International has redomiciled from the British Virgin Islands to El Salvador and now operates as Tether International, S.A. de C.V.; its token terms nevertheless continue to use BVI governing law.4 The frozen assets were USDT tokens on the TRON blockchain. The freeze occurred before judicial review, before OFAC’s public SDN List update the following day and before any constructive trust or judgment was imposed.

This is a different legal architecture from anything the US has previously deployed against Bank Markazi. In Peterson, Justice Ginsburg’s majority opinion (6-2, Roberts CJ and Sotomayor J dissenting) held that Congress had not invaded the judicial power by directing the disposition of specific assets in pending litigation.5 The dissent warned that the legislative branch had effectively decided a particular case. The 2026 mechanism does not even reach those waters. There is no statute directing courts. There is no judgment. There is a private issuer freezing tokens in coordination with OFAC and US law enforcement.

The GENIUS Act formalises the issuer-freeze model

The 10 April 2026 FinCEN-OFAC joint Notice of Proposed Rulemaking, issued under the GENIUS Act (S.1582, 119th Congress), would require every permitted payment stablecoin issuer to maintain technical capability to block, freeze and reject impermissible transactions, including transactions occurring on secondary markets via smart contracts.6 The proposed rule treats issuers as financial institutions under the Bank Secrecy Act, mandates risk assessments, internal controls, training and audit, and obliges issuers to comply with lawful orders. Comments close on 9 June 2026.

Once finalised, the rule will convert issuer-level freeze capability from a voluntary practice into a condition of operating as a US-permitted payment stablecoin issuer. Issuers of USDC, PYUSD and other payment stablecoins will be subject to that obligation where they operate through a PPSI structure.

The cross-jurisdictional gap

The UK and EU sanctions regimes already apply to cryptoassets. The cross-jurisdictional gap is more specific: neither the UK stablecoin regime nor MiCA yet appears to impose a GENIUS-style issuer-level obligation to maintain native technical capability to block, freeze and reject secondary-market stablecoin transactions through smart-contract control.

In the UK, FSMA 2023 and the developing FCA/Bank of England regimes focus principally on authorisation, backing assets, redemption, safeguarding, prudential requirements, operational resilience and systemic payment risk. UK sanctions law may still require cryptoassets to be frozen where they are funds or economic resources of a designated person, but that is not the same as requiring every stablecoin issuer to maintain smart-contract-level blocking infrastructure.7

In the EU, MiCA imposes authorisation, reserve, redemption, governance and disclosure obligations on asset-referenced tokens and e-money tokens. Sanctions enforcement continues to operate through EU asset-freeze regulations and Member State enforcement structures. There is no MiCA equivalent of the GENIUS Act’s PPSI-level technical freeze capability.8

The result is that, for a period, US regulation reaches more directly into the issuer-level technical architecture of stablecoin control than the equivalent UK or EU stablecoin regimes do. For Iranian, Russian and other sanctioned counterparties, the practical implication is that USDT and USDC are exposed to Treasury-driven issuer action in a way that GBP or EUR denominated stablecoins issued solely under UK or EU regimes may not be.

Forward implication

The Bank Markazi action will be cited and copied. It is a clean operational precedent for OFAC: identify wallets, coordinate with the issuer, freeze the tokens and then add the addresses publicly to the SDN entry. Practitioners advising sanctioned counterparties, or counterparties exposed to sanctions risk, should assume that stablecoin balances issued by entities with native freeze functionality and material US sanctions exposure may be frozen at issuer level before any judicial process. For US-permitted payment stablecoin issuers, the proposed GENIUS Act rule would turn that capability into a regulatory condition of issuance.

The Bank Markazi v Peterson architecture has not been overturned. It has been bypassed.

In November 2025, the United States Department of Justice announced a proposed consent judgment with RealPage, Inc., a software company whose revenue management tools operated across residential housing markets by aggregating tenant and rental data from thousands of competing property managers.

The proposed consent judgment is one of the clearest federal templates yet for regulating the architecture of pricing software. It treats algorithmic pricing tools not merely as passive decision-support systems, but as potential instruments for unlawful information sharing and price alignment, advancing the enforcement theory that software can function as an instrument of unlawful price coordination even where no traditional human “meeting of minds” is pleaded in conventional cartel terms.

This marks a regulatory turning point in what commentators increasingly describe as the “algorithmic cartel” and what enforcers more often frame as algorithmic coordination, information sharing and pricing alignment - a phenomenon in which automated systems achieve convergent, supra-competitive market outcomes by processing competitively sensitive data in real-time without explicit human agreement.1 What makes this development urgent is not the novelty of price-fixing itself, but the speed and opacity at which coordination now occurs. A landlord accepting a pricing recommendation may not realise that the system is being treated by enforcers as part of an alleged anticompetitive information-sharing architecture. An autonomous system designed to maximise profit may independently learn that coordination produces better returns than competition. These scenarios challenge the foundational legal concept that cartels require a “meeting of minds.” The data-hub version of the problem is no longer theoretical; the fully autonomous version is now close enough to the market to worry regulators.

The RealPage Proposed Consent Judgment: Architectural Restrictions on Algorithmic Influence

The proposed DOJ-RealPage settlement addresses coordination through three distinct mechanisms: the age of competitor data used to train pricing models, the geographic scope of that data and the architecture of software choice itself. At the centre of the DOJ’s complaint was RealPage’s YieldStar and AI Revenue Management (AIRM) platforms, which functioned as a “melting pot” of confidential competitor information.2,3 By pooling granular, non-public lease data from tens of thousands of properties (including actual transactional prices, future occupancy levels and specific lease terms), RealPage’s models generated pricing recommendations that maximised total industry revenue rather than reflecting the independent competitive interests of individual landlords.4,5

The innovation of the settlement lies in its distinct treatment of runtime operation versus model training. The DOJ recognised that the most immediate collusion risk occurs when an algorithm uses a competitor’s real-time, non-public data to adjust prices in the current market.6 The proposed judgment would prohibit RealPage from utilising any competitively sensitive information (CSI) from unaffiliated properties during runtime operations of its revenue management tools.2,6 This restriction would ensure that a landlord’s daily pricing updates are based either on their own proprietary data or on information that is “readily accessible to the general public”.6

Regulators took a different approach toward model training, acknowledging that AI systems require access to historical datasets to maintain predictive accuracy and improve market efficiency.2,7 The proposed judgment would permit RealPage to train its models on non-public competitor data, but only if that data is at least 12 months old.2,6 This temporal buffer neutralises the strategic potency of the data while preserving its utility for identifying long-term cyclical trends in supply and demand.

The proposed judgment would also bar RealPage from using models that determine geographic effects narrower than state level.2,6 That requirement is broader than the rental markets alleged in the complaint and is designed to reduce the risk of hyper-local price alignment.8,9 By preventing the algorithm from learning and enforcing localised pricing patterns, the proposed judgment aims to dilute the risk of tacit coordination within a specific neighbourhood or census block while still allowing the software to account for regional economic shifts.

Choice Architecture and the Coercion of Compliance

The DOJ’s investigation revealed that the effectiveness of the RealPage system relied not just on the data itself, but on the choice architecture designed to suppress independent decision-making by property managers. The complaint alleged that RealPage’s software was engineered to make accepting price recommendations a frictionless process while making overrides difficult and time-consuming.3,5 Property managers could “bulk accept” multiple recommendations but were required to provide “specific business commentary” for every rejection.3,5 These justifications were reviewed by RealPage “pricing advisors” and could be escalated to regional managers. This created a psychological and bureaucratic deterrent against competitive pricing.3,5

The proposed consent judgment would require RealPage to eliminate these coercive design features. Software settings, such as “Auto-Accept” or “Governor” features (which were alleged to favour price increases over decreases), would need to be symmetrical and require active, manual configuration by the user.6 This provision establishes a core principle emerging in modern antitrust enforcement: algorithmic tools must serve as aids to independent human judgment rather than as automated mandates for market alignment.10,11

The Agri Stats Precedent: Benchmarking as a Coordination Hub

While the RealPage case focused on the technology of AI itself, the Agri Stats litigation demonstrates how traditional data-benchmarking services have evolved into instruments of algorithmic coordination. Agri Stats sits alongside RealPage as a broader information-exchange precedent, although the procedural posture differs. Agri Stats collected and distributed detailed financial and production data among the leading processors of chicken, pork and turkey.12,13 The DOJ’s case alleges that Agri Stats’ reports enabled meat processors to use competitively sensitive information to stabilise or increase prices and reduce output across the US protein industry.12,14

Separately, in the private wage-suppression litigation, Agri Stats settlements required changes to labour-related reporting, including removal or redaction of plant-level labour data, preventing meatpackers from monitoring the specific compensation strategies of their rivals.13,15 This redact-at-source requirement targets the transparency problem that facilitates coordination in oligopolistic markets.

The Withdrawal of Regulatory Safe Harbours

The Agri Stats and RealPage cases must be viewed against a broader regulatory shift: the withdrawal of the 1996 and 2000 antitrust safety zones for information sharing.13,16 For decades, firms operated under the assumption that sharing data was legal if it was managed by a third party, involved at least five participants, and was at least three months old.2,13 The withdrawal began earlier than the current algorithmic-pricing cases. DOJ withdrew the health-care information-sharing policy statements in February 2023, the FTC followed in July 2023, and both agencies withdrew the 2000 Collaboration Guidelines in December 2024.7,16 The rationale was consistent: existing safe harbours were overly permissive and failed to account for the ability of modern algorithms to de-anonymise aggregated data and exploit even three-month-old information in real-time.

The Strategic Uncertainty Crisis in Hospitality: The CMA Hotel Investigation

On 24 February 2026, announced publicly on 2 March 2026, the United Kingdom’s Competition and Markets Authority (CMA) launched a Competition Act 1998 investigation into three global hotel giants (Hilton, Marriott and IHG) alongside the data analytics provider STR, owned by CoStar Group.17,18 The investigation centres on the suspected sharing of competitively sensitive information through STR’s benchmarking tools.17,19 STR monitors over 90,000 hotels globally and provides the industry with reports on occupancy, average daily rates (ADR), and revenue per available room (RevPAR).18,20

The CMA’s theory of harm focuses on reduction of strategic uncertainty. In a competitive market, a firm’s inability to predict its rival’s pricing and capacity decisions drives it to lower prices or improve service.17,20 The CMA is investigating whether the use of a common data-services provider reduced strategic uncertainty between competing hotel chains by enabling the exchange of competitively sensitive information. The CMA has not reached a view on whether there is sufficient evidence of infringement.17 This case matters because it demonstrates how traditional benchmarking platforms may transition into hubs for algorithmic coordination in highly dynamic service markets.

Agentic AI and the Risk of Autonomous Collusion

The hotel investigation sits against a broader CMA concern about pricing algorithms, AI systems and the possibility that automated tools may facilitate collusion or reduce strategic uncertainty.21 The CMA warned that when competing firms deploy agents programmed to maximise profit, these agents may independently discover that coordination produces returns superior to aggressive competition.15,21 This autonomous collusion presents a major enforcement challenge because it can occur without any direct communication between the human owners of the agents.15,21

Transatlantic Divergence: The Doctrinal Chasm

The central legal debate in both US and EU jurisdictions is whether independent adoption of the same pricing algorithm constitutes a violation of antitrust laws. This debate exposes a growing gap between established law and the realities of digital markets.1,2

US Jurisprudence: The “Agreement” Requirement vs. Plus Factors

Under Section 1 of the Sherman Act, the US government must prove the existence of a “contract, combination, or conspiracy”.10 In the Gibson v. Cendyn case (August 2025), the Ninth Circuit affirmed the dismissal of a class action against several hotel chains, ruling that simply subscribing to the same pricing software was insufficient to establish an agreement.8 The court described this as “consciously parallel conduct,” which is not illegal unless accompanied by “plus factors” that suggest a meeting of minds.8

Other US courts have adopted a broader view. In Duffy v. Yardi (December 2024), a district court in Washington denied a motion to dismiss, finding that the software provider’s marketing (which promised to help landlords “maximize rents across the industry”) functioned as an “invitation to conspire”.8,12 By accepting this invitation and providing their commercially sensitive data to the melting pot, the landlords effectively entered into an unlawful horizontal agreement.8,12

EU Framework: The “Concerted Practice” and Article 101 TFEU

Article 101(1) TFEU prohibits not only agreements but also “concerted practices” (a broader category that encompasses any direct or indirect contact between competitors intended to influence market behaviour).2,22 This framework is inherently more adaptable to algorithmic coordination. Linsey McCallum, the Deputy Director General of the European Commission’s competition arm, stated in July 2025 that the Commission had identified red flags in multiple confidential investigations where algorithms facilitate or monitor coordination between competitors.15,23

In the EU, two scenarios are at the forefront of the debate. The “Predictable Agent” refers to algorithms that respond to market signals in a way that leads to tacit coordination. The “Digital Eye” describes autonomous systems that independently learn that collusion is the most profitable strategy.1,2 European enforcers are moving toward an attribution-and-control model: undertakings cannot avoid liability merely because pricing decisions are mediated through software that they chose, configured, supplied with data or failed to supervise.24

Sectoral Enforcement in the European Economic Area

While the European Commission pursues large-scale investigations, national competition authorities are taking the lead in specific sectors. These investigations provide practical insights into how Article 101 is being applied to algorithmic cartels.24

Poland’s UOKiK: Banking and Pharmaceuticals

In September 2025, the President of the Polish Office for Competition and Consumer Protection (UOKiK) confirmed two major investigations into algorithmic pricing. The banking investigation reportedly involves several Polish banks that allegedly utilised algorithms fed by data from the country’s largest credit risk database and their own non-public internal information to coordinate the pricing of consumer loans and mortgages.15 The pharmaceutical wholesale probe reportedly concerns three major wholesalers controlling 80% of the market, who allegedly used IT systems to exchange commercially sensitive information on drug margins, prices, and volumes sold through affiliated pharmacies.12,25 This should be distinguished from earlier UOKiK proceedings concerning pharmaceutical wholesalers and software providers, which also involved alleged exchanges of commercially sensitive information but arose in a separate procedural context.25

These cases matter because they target both the spokes (the banks and wholesalers) and the hub (the software providers), reflecting a regulatory consensus that the facilitators of algorithmic collusion share in the liability.26

Legislative Frontiers: Bridging the Proof Gap

The difficulty of proving a meeting of minds in the era of deep learning has prompted legislative action. These bills seek to change the presumptions of antitrust law, shifting the burden of proof from the regulator to the firm.

The Preventing Algorithmic Collusion Act was reintroduced in January 2025 by Senator Amy Klobuchar as Senate Bill 232. This legislation aimed to modernise the Sherman Act by creating a legal presumption of collusion for certain uses of pricing algorithms.2 The bill would have prohibited the use of algorithms that rely on non-public competitor data and mandated algorithmic audits.2,10 The bill was introduced on 23 January 2025 and referred to the Senate Judiciary Committee. As at publication, it had not advanced beyond that stage.27 Industry objections have focused on innovation, overbreadth and tension with conventional rule-of-reason analysis.2

While federal algorithmic-collusion legislation has not advanced, California and New York emerged as the primary antitrust frontier in 2025 and 2026.28,29 California AB 325, approved in October 2025 and operative from 1 January 2026, amends the Cartwright Act to prohibit the use or distribution of a common pricing algorithm as part of an unlawful contract, combination or conspiracy in restraint of trade. It also prohibits using or distributing such an algorithm where a person coerces another to adopt a recommended price or commercial term, and lowers the pleading threshold by requiring only plausible allegations of conspiracy rather than facts excluding independent action.28,30 AB 325 will nevertheless make due diligence practically essential, because businesses will need to understand whether their tools use competitor data and whether any feature could be characterised as coercing adoption of recommended prices.

New York moved on two tracks: a housing-specific restriction on algorithmic rent-setting and a separate consumer-facing disclosure law for algorithmic pricing based on personal data. AB A1417B bans landlords from using rent-setting software that utilises non-public competitor data.15,28 The separate Algorithmic Pricing Disclosure Act, now in effect, requires most businesses using personalised algorithmic pricing based on consumer personal data to display a clear disclosure near the price.28

Conceptual Models of Algorithmic Collusion

To effectively regulate algorithmic pricing, authorities have identified three distinct scenarios of harm, each presenting unique evidentiary challenges.

The first scenario is the Automated Traditional Cartel. Here the algorithm is merely a tool used to implement an existing human agreement. Competitors reach a meeting of minds via email or meeting and then program their repricing software to enforce the arrangement.1 This occurred in the 2016 UK Posters case, where sellers on Amazon used software to ensure they never undercut each other.1,23 Existing Section 1/Article 101 doctrine easily captures this scenario because criminal intent remains human.

The second scenario is Hub-and-Spoke Coordination. Multiple competitors delegate their pricing decisions to the same common algorithm (the hub). Even if competitors do not communicate directly, they knowingly enter into a shared system that pools their sensitive data and provides aligned recommendations.1,31 The RealPage and Agri Stats cases exemplify this model. The legal challenge lies in proving that the spokes (competitors) understood the collaborative nature of the hub.15,26

The third scenario is the “Digital Eye.” AI systems trained to maximise profits independently discover that coordination produces better outcomes than competition.1 Through reinforcement learning, the software learns to signal its intention to maintain high prices and to punish any rival that cheats by lowering them.2,24 Because there is no human agreement and no shared hub, this scenario falls into a legal grey zone where traditional doctrines of coordinated conduct are difficult to apply.1,2

Economic Implications: The End of Strategic Uncertainty

Algorithmic pricing has altered the game theory of market competition. Economic theory suggests that price transparency benefits consumers. In digital markets, however, this transparency is often asymmetrical.1,23

Algorithms allow firms to monitor rivals with extreme speed and granularity. This high-frequency monitoring reduces reaction time to zero, meaning any firm attempting to gain market share by lowering prices is instantly matched.1,15 When price-cutting becomes futile, competition shifts from a race to the bottom to an equilibrium at supra-competitive levels. The allegations against RealPage’s Governor feature illustrate this price stickiness in the upward direction.3,6

The RealPage settlement also highlighted how software design can nudge users toward collusive outcomes. By creating an environment where diverging from a recommendation requires high administrative effort, software providers effectively strip human agents of their independent competitive instincts.3,5 Even if a firm has no intent to collude, its participation in a shared pricing ecosystem can lead to anti-competitive market effects through algorithmic coercion.11,28

The Future of Enforcement: From Hindsight to Oversight

The settlements of late 2025 and early 2026 indicate a permanent shift in how antitrust authorities view technology. The wait-and-see approach of the early 2020s has been replaced by aggressive, proactive enforcement.21,23

Enforcers now demand that companies conduct internal audits of their pricing tools. The CMA and EC have emphasised that businesses must understand the sources of underlying training data and ensure that their software’s objectives are not promoting aligned or elevated pricing.15,26 There is growing academic and regulatory push to apply the precautionary principle to algorithmic markets.34 This approach argues that because of unknown unknowns associated with deep learning and the speed at which digital markets tip into monopolies, regulators should intervene early to prevent harm rather than waiting for empirical proof of an antitrust violation.34 The CMA hotel case is not a market investigation; it is a Competition Act 1998, Chapter I investigation. But it sits within a broader UK move from passive observation toward earlier scrutiny of data hubs, pricing tools and algorithmic systems.15

Conclusion: Mapping the Unresolved Territory

The algorithmic cartel exposes fundamental gaps in 20th-century competition law. From human-mediated agreements to data-mediated convergence, the regulatory environment has shifted dramatically. The Sherman Act and Article 101 TFEU are being stretched in real-time to cover automated coordination. RealPage and Agri Stats demonstrate that regulators are finding ways to adapt legacy doctrines to algorithmic reality.

If entered, the RealPage model - defined by 12-month data ageing, statewide granularity restrictions and mandatory symmetry in software settings - is likely to become an influential template for algorithmic-pricing compliance.2,6 Investigations in the UK (hotels) and Poland (pharmaceuticals) signal that regulators will no longer tolerate third-party data hubs that eliminate strategic uncertainty.

Yet the law remains uncertain on the question that will define enforcement in the next phase. The Digital Eye scenario involves autonomous systems that collude without human direction or shared data through independent learning that coordination maximises profit - a possibility that falls into a regulatory grey zone. No existing doctrine cleanly captures it. The hard case is narrower but more profound: autonomous tacit coordination without contact, without a shared hub, without competitor data exchange and without an attributable instruction to collude. That is where the gap between economic harm and legal doctrine remains most exposed. Section 1 of the Sherman Act requires an agreement; Article 101 TFEU can capture indirect contact and concerted practices and liability can attach to conduct implemented through tools or agents. But an autonomous system that learns to collude through reinforcement learning presents none of these features. It may produce identical outcomes to a cartel, with identical consumer harm, but the legal tools designed to address cartels do not fit. This gap between harm and remedy is the unresolved question that will occupy enforcement agencies and courts throughout 2026 and beyond. For businesses, the path forward requires moving from algorithmic blind faith to architectural compliance - designing systems that compete rather than coordinate, using public data instead of competitor CSI, with responsibility for competitive behaviour now resting in the code itself.

References

1 Algorithmic Collusion: Corporate Accountability and the Application of Art. 101 TFEU, https://www.europeanpapers.eu/europeanforum/algorithmic-collusion-corporate-accountability-application-art-101-tfeu

2 Algorithmic Tacit Collusion: Addressing the Gaps in Article 101(1)(a) of the TFEU - Knight-Georgetown Institute, https://kgi.georgetown.edu/wp-content/uploads/2026/01/Algorithmic-Tacit-Collusion_Brambilla_17.pdf

3 Proposed DOJ settlement provides guidance on use of competitive information in algorithmic pricing tools - Hogan Lovells, https://www.hoganlovells.com/en/publications/proposed-doj-settlement-provides-guidance-on-use-of-competitive-information

4 United States of America et al. v. RealPage, Inc. et al.; Proposed Final Judgment and Competitive Impact Statement - Federal Register, https://www.federalregister.gov/documents/2025/12/05/2025-21966/united-states-of-america-et-al-v-realpage-inc-et-al-proposed-final-judgment-and-competitive-impact

5 Last Year’s Rent: RealPage Reaches Settlement Agreement with the Department of Justice in Algorithmic Pricing Case | Mintz, https://www.mintz.com/insights-center/viewpoints/2191/2025-12-01-last-years-rent-realpage-reaches-settlement-agreement

6 United States of America et al. v. RealPage, Inc. et al. Proposed Final Judgment and Competitive Impact Statement - Federal Register, https://www.federalregister.gov/documents/2026/01/21/2026-01009/united-states-of-america-et-al-v-realpage-inc-et-al-proposed-final-judgment-and-competitive-impact

7 Practical Takeaways From the DOJ’s Algorithmic Pricing Settlement, https://www.paulweiss.com/insights/client-memos/practical-takeaways-from-the-doj-s-algorithmic-pricing-settlement

8 Ninth Circuit Clarifies Antitrust Implications of Algorithmic Pricing, https://www.arnoldporter.com/en/perspectives/advisories/2025/08/antitrust-implications-of-algorithmic-pricing

9 DOJ Press Release: Justice Department Requires RealPage to End Sharing of Competitively Sensitive Information and Redesign Key Software Tools, 24 November 2025, https://www.justice.gov/opa/pr/justice-department-requires-realpage-end-sharing-competitively-sensitive-information-and

10 FTC and DOJ Statement of Interest, Cornish-Adebiyi v Caesars Entertainment, filed 28 March 2024; FTC press release, “FTC and DOJ File Statement of Interest in Hotel Room Algorithmic Price-Fixing Case”, 28 March 2024, https://www.justice.gov/archives/opa/pr/justice-department-and-federal-trade-commission-file-statement-interest-hotel-room

11 Corporate-Tech Landlordism REVISED 19Aug2025 - Stanford Law School, https://law.stanford.edu/wp-content/uploads/2025/07/Corporate-Tech-Landlordism-REVISED-19Aug2025.pdf

12 Gig Platforms as Hub-and-Spoke Arrangements and Algorithmic Pricing: A Comparative EU-US Antitrust Analysis - OpenEdition Books, https://books.openedition.org/putc/15512

13 Settlements Reached with Agri Stats in Broilers, Turkey, Pork, https://www.hbsslaw.com/press/pork-antitrust/settlements-reached-with-agri-stats-in-broilers-turkey-pork-antitrust-suits-over-price-fixing-allegations

14 Expect more turmoil and change for the 2026 ag sector - Investigate Midwest, https://investigatemidwest.org/2026/01/08/expect-more-turmoil-and-change-for-the-2026-ag-sector/

15 CMA Case Page: Suspected sharing of competitively sensitive information by Hilton, IHG, Marriott and STR (CoStar), CA98/01/2026, https://www.gov.uk/cma-cases/suspected-sharing-of-competitively-sensitive-information-by-hilton-ihg-marriott-and-str-costar

16 DOJ Press Release: Justice Department Withdraws Outdated Enforcement Policy Statements, February 2023, https://www.justice.gov/archives/opa/pr/justice-department-withdraws-outdated-enforcement-policy-statements; FTC withdrawal of healthcare antitrust policy statements, July 2023; DOJ/FTC joint withdrawal of 2000 Collaboration Guidelines, December 2024

17 UK Watchdog Probes Data-Sharing Among Hotel Giants - PYMNTS.com, https://www.pymnts.com/data/2026/uk-watchdog-probes-data-sharing-among-hotel-giants/

18 Hilton, IHG and Marriott under CMA investigation for information sharing - The Caterer, https://www.thecaterer.com/news/hilton-ihg-and-marriott-under-cma-investigation-for-information-sharing

19 UK CMA Investigates Hotels, CoStar Over Data Sharing - Asian Hospitality, https://www.asianhospitality.com/uk-cma-hotel-data-sharing-probe/

20 UK’s CMA launches investigation into Hilton, IHG and Marriott, https://www.businesstravelnewseurope.com/Accommodation/UK-s-CMA-launches-investigation-into-Hilton-IHG-and-Marriott

21 Europe steps up antitrust enforcement against algorithmic pricing | Loyens & Loeff, https://www.loyensloeff.com/insights/news--events/news/europe-steps-up-antitrust-enforcement-against-algorithmic-pricing/

22 EU rules on concerted practices and agreements between companies - EUR-Lex, https://eur-lex.europa.eu/EN/legal-content/summary/eu-rules-on-concerted-practices-and-agreements-between-companies.html

23 Recent Algorithmic Pricing Developments in the UK and the EU, https://perkinscoie.com/insights/update/recent-algorithmic-pricing-developments-uk-eu

24 The Misleading Consequences of Comparing Algorithmic and Tacit Collusion - European Papers, https://www.europeanpapers.eu/system/files/pdf_version/EP_eJ_2021_2_6_Articles_Luca_Calzolari_00519.pdf

25 Bird & Bird, UOKiK probes software-enabled exchange of strategic information between pharmaceutical wholesalers, 2022 (note: this relates to an earlier proceeding, not the September 2025 algorithmic-pricing investigations), https://www.twobirds.com/en/insights/2022/poland/uokik-probes-software-enabled-exchange-of-strategic-information

26 Algorithmic Pricing Emerges as Enforcement Priority for EU & UK Antitrust Regulators, https://www.morganlewis.com/pubs/2025/10/algorithmic-pricing-emerges-as-enforcement-priority-for-eu-and-uk-antitrust-regulators

27 The Preventing Algorithmic Collusion Act: Strike two - DLA Piper, https://www.dlapiper.com/insights/publications/2025/02/the-preventing-algorithmic-collusion-act-2025

28 New laws regulating algorithmic pricing enacted in New York and California | Davis Polk, https://www.davispolk.com/insights/client-update/new-laws-regulating-algorithmic-pricing-enacted-new-york-and-california

29 On the continuing relevance of state antitrust enforcement in the US - Herbert Smith Freehills, https://www.hsfkramer.com/insights/2026-01/on-the-continuing-relevance-of-state-antitrust-enforcement-in-the-us

30 A Year at the Justice Department’s Antitrust Division | The Regulatory Review, https://www.theregreview.org/2026/03/25/slater-a-year-at-the-antitrust-division/

31 Gig Platforms as Hub-and-Spoke Arrangements and Algorithmic Pricing: A Comparative EU-US Antitrust Analysis - OpenEdition Books, https://books.openedition.org/putc/15512

32 Algorithmic Pricing and AI-Powered Evidence Avoidance: Competition Law Risks and Compliance Strategies | Goodwin - JDSupra, https://www.jdsupra.com/legalnews/algorithmic-pricing-and-ai-powered-1056402/

33 Adapting to and Getting Ahead of Changes in Antitrust and Other Regulatory Demands in 2025 and Beyond - Redgrave LLP, https://www.redgravellp.com/publication/adapting-to-and-getting-ahead-of-changes-in-antitrust-and-other-regulatory-demands-in-2025-and-beyond

34 Synthetic Futures and Competition Law: Towards the Emergence of Precautionary Principle-Minded Approaches - University College London, https://www.ucl.ac.uk/laws/sites/laws/files/cles-6-2024_1.pdf

The signal before the fairness hearing

The claims process in Bartz v. Anthropic PBC closed at a claim rate of 91.3 per cent. Of 482,460 eligible works in the class, 440,490 have been claimed.1 The mid-point checkpoint on 19 March 2026 sat at 54 per cent. The figure climbed thirty-seven percentage points in the final eleven days before the 30 March claim deadline.

That is not a normal claims-made settlement. The often-cited benchmark for claims-made consumer class settlements is much lower: the FTC’s 2019 study reported a median calculated claims rate of 9 per cent, excluding cases where the calculated rate exceeded 100 per cent.2 Bartz is therefore an outlier signal. It suggests that notice reached the people it needed to reach, that class members saw meaningful value in claiming, and that the practical force of adequacy objections is reduced.

The fairness hearing is no formality. Judge Martínez-Olguín has moved the hearing from 23 April to 14 May 2026 and has permitted timely objectors to be heard.3 The arithmetic, though, is already doing heavy work in the plaintiffs’ favour.

The distinction that forced the deal

The piracy theory, not the training theory, is what Anthropic is paying $1.5 billion to settle. The distinction matters.

On 23 June 2025, Judge Alsup granted partial summary judgment holding that the copies used to train Claude and its precursors were fair use, and that digitising purchased print books was also fair use. He refused, however, to treat Anthropic’s pirated central-library copies as protected by the same analysis.4 The reasoning was analytical. Training on copyrighted texts to synthesise a generative model, on the court’s view, is transformative in the manner a human reader is transformative when they read widely to learn to write.

Library Genesis and Pirate Library Mirror were not merely acquisition channels for training. On Judge Alsup’s analysis, they were the source of a permanent, general-purpose corpus of infringing files retained on the company’s servers. The court treated that corpus as a separate use requiring its own justification. Statutory damages for wilful infringement run up to $150,000 per work.5 For a class of roughly half a million registered works, the theoretical exposure ran into the tens of billions. The $1.5 billion settlement is priced against that retention exposure, not against the training use itself.

Anthropic settled because the piracy and central-library issue created live, class-wide statutory damages exposure, not because Judge Alsup had rejected fair use for training as such. The implication for the broader AI copyright docket is structural.

The settlement in numbers

The gross fund is $1.5 billion. Requested deductions total roughly $208.6 million, producing a net distribution fund of approximately $1.29 billion.6 The base payout per claimed work, before interest accrual, comes to $2,931.62. Claims will be adjudicated and distributions calculated by approximately 11 June 2026, with payment commencement expected no earlier than late autumn 2026.7

Anthropic has placed $300 million into escrow with further tranches due after final approval, before 25 September 2026 and before 25 September 2027.8 The staged funding is standard in settlements of this magnitude and is one reason final approval is not a formality. The court needs to be satisfied that the funding schedule is secured against Anthropic’s balance sheet and against foreseeable corporate change. It is unusual for a copyright defendant to pay on this scale over this period, and the court will want the tranche mechanics scrutinised in open court.

The 12.5 per cent fee request

Class counsel have moved the attorney fee request from 25 per cent of the fund to 12.5 per cent, equal to $187.5 million.9 Expenses add $2.78 million, a cost reserve of $18.22 million sits alongside and service awards of $150,000 are distributed across the three named plaintiffs at $50,000 each.10

The fee arithmetic is the part of the settlement most likely to draw favourable attention at the fairness hearing. Class action fee percentages often sit materially higher than 12.5 per cent: the Ninth Circuit’s common-fund benchmark is commonly stated as 25 per cent, and Authors Guild describes class action fees as ranging around 30 per cent. A 12.5 per cent request is uncommonly low in absolute terms and markedly low in relative terms against a fund of this size. The step-down reads as a strategic move to pre-empt a common adequacy objection and to narrow the surface area available to objectors who want to attack the economics.

The unsealed objections

The court ordered a substantial block of objections unsealed in early April 2026, including dockets 425, 539-546, 549-552, 564-569, 584-585, 588-589 and 593-612.11 Authors Alliance later noted that, as of 20 April 2026, several of the listed objections still appeared to remain sealed, including dockets 544, 596, 598, 600, 601 and 602. Named objectors include Professor Lea Victoria Bishop and an objector identified as Esquivel.

The grounds fall into five families. First, scope. The class as certified excludes foreign and non-US-registered works. Objectors put the excluded population at more than two million books. Second, quantum. Statutory damages run up to $150,000 per wilfully infringed work. A settlement offer averaging roughly $3,000 per work represents a reduction of some 98 per cent against the statutory ceiling. Third, registration. Group registrations are being treated as single compensable works even where they collapse forty or more titles. Fourth, split. The allocation mechanism permits publishers to claim approximately half of the per-title payment before the author sees any of it. Fifth, precedent. Some objectors argue the settlement sets a cheap liability template that the next defendant will try to replicate.

None of those objections is obviously frivolous. The first is a structural line-drawing choice that reflects the plaintiffs’ registration-based theory of infringement rather than a settlement compromise. The second and third go to adequacy and are the familiar territory of fairness review. The fourth reflects longstanding author-publisher tension over book-revenue allocation and the fifth is policy rather than law. Judge Martínez-Olguín will weigh each on the Rule 23(e) adequacy standard, but the 91.3 per cent claim rate cuts against all of them as a matter of record. A class that overwhelmingly elects to claim has, on a practical reading, told the court that the deal was worth taking.

Kadrey, Meta and the limits of the parallel

Bartz leaves Judge Alsup’s fair use holding on training untouched. Kadrey v. Meta Platforms, Inc. reached a similar bottom-line result on training, but on a materially different and heavily caveated record.12 Judge Chhabria granted Meta summary judgment on fair use, while making clear that the ruling did not establish a general rule that Meta’s use of copyrighted works to train language models was lawful. It reflected the plaintiffs’ failure to develop the evidential record needed to defeat fair use.

The safer proposition is therefore narrower. On the current Northern District of California authorities, training remains a fair use defence available to model developers. It is not a settled rule. The next result will depend on the record: provenance, market harm, output evidence and whether plaintiffs can move beyond the evidential gaps that weakened Kadrey.

The route that remains most clearly open is the one Anthropic lost on. Show that the model was trained on, or built on top of, data the defendant did not lawfully acquire, and pursue retention and reproduction theories rather than the training theory in isolation. That is the litigation template visible in the dockets now moving through 2026 and 2027. It runs through discovery into acquisition provenance rather than through argument about the fair use factors alone.

OpenAI, Midjourney and the live dockets

The next wave of cases does not mirror Bartz on the facts. The consolidated OpenAI copyright litigation in the Southern District of New York, before Judge Sidney Stein, now includes a 5 January 2026 order affirming Magistrate Judge Ona Wang’s direction that OpenAI produce the full 20 million-log ChatGPT sample.13 The Disney, NBCUniversal and Warner Bros. Discovery actions against Midjourney and MiniMax, commenced between June and September 2025, raise output infringement on a visual-corpus theory that did not feature in Bartz at all.14 Those cases go to whether the model’s outputs reproduce copyrighted expression in a way the training cases did not need to decide.

The Bartz settlement does not dispose of any of those theories. It establishes two things only. It confirms that where a defendant lifted pirated corpora from LibGen or PiLiMi, liability is real and pricing can reach roughly $2,931 per registered work. It confirms, by not disturbing Judge Alsup’s holding, that on the current Northern District of California authorities training on legally acquired material remains a fair use defence available to the model developer. Everything beyond those two points is open.

The fork

The analytical question for 2026 and 2027 is whether follow-on plaintiffs can demonstrate that legal acquisition was a cover for scraped or pirated data. If they can, $2,931.62 per registered work becomes a reference point for damages discussions, not the ceiling. If they cannot, the Bartz and Kadrey defendants’ record on training holds on its narrow facts, and the plaintiffs’ bar will need to pivot to output infringement theories of the kind now pending in the Midjourney and MiniMax dockets.

That is the fork the AI copyright bar now faces. The 91.3 per cent claim rate tells the court that eligible rightsholders overwhelmingly elected to participate in the settlement economics. It does not tell future courts what to do with the next model. Future courts will ask where the training data came from, whether any of it was pirated and whether any of the outputs reproduce protectable expression. None of those questions has a settled answer. All of them are the subject of active litigation.

The settlement, if approved after the 14 May hearing, will draw a line under Anthropic’s class-wide piracy exposure for the covered works. It will not draw a line under the AI copyright question. The industry will read $1.5 billion as the price of piracy. The plaintiffs’ bar will read $2,931.62 as a reference point for registered works. Both readings are right and both are partial. The rest of the question is in discovery.

Footnotes

1. Authors Guild, Anthropic Settlement Update: 91.3 Percent of Books Claimed (17 April 2026) https://authorsguild.org/news/anthropic-settlement-update-91-percent-of-books-claimed/

2. Federal Trade Commission, Consumers and Class Actions: A Retrospective and Analysis of Settlement Campaigns (2019) (median calculated claims rate of 9 per cent across claims-made consumer class settlements, excluding cases where the calculated rate exceeded 100 per cent).

3. Authors Alliance, Bartz v. Anthropic Settlement Update: New Date and Time for the Fairness Hearing and Unsealed Objections (14 April 2026, updated 20 April 2026) https://www.authorsalliance.org/2026/04/14/bartz-v-anthropic-settlement-update-new-date-and-time-for-the-fairness-hearing-you-can-join-online-and-unsealed-objections/

4. Bartz v. Anthropic PBC, No. 3:24-cv-05417 (N.D. Cal.), Order on Partial Summary Judgment (23 June 2025) (Alsup J).

5. 17 U.S.C. § 504(c)(2).

6. Authors Guild (n 1).

7. ibid.

8. ibid.

9. PYMNTS, Anthropic Copyright Settlement Lawyers Cut Fee Request to $187.5 Million (20 March 2026) https://www.pymnts.com/cpi-posts/anthropic-copyright-settlement-lawyers-cut-fee-request-to-187-5-million/

10. Authors Guild (n 1).

11. Authors Alliance (n 3).

12. Kadrey v. Meta Platforms, Inc., No. 3:23-cv-03417 (N.D. Cal.), Order on Cross-Motions for Partial Summary Judgment (2025) (Chhabria J). The court granted Meta summary judgment on fair use while making clear that the ruling did not establish a general rule that Meta’s use of copyrighted works to train language models was lawful and reflected the plaintiffs’ failure to develop the necessary evidential record.

13. Reuters, OpenAI loses fight to keep ChatGPT logs secret in copyright case (3 December 2025) (reporting Magistrate Judge Wang’s December 2025 production order); Jones Walker, OpenAI Loses Privacy Gambit: 20 Million ChatGPT Logs Likely Headed to Copyright Plaintiffs (January 2026) (reporting Judge Stein’s affirmance of 5 January 2026); In re: OpenAI, Inc. Copyright Infringement Litigation, S.D.N.Y.

14. MBHB, AI News Roundup: Anthropic settles copyright infringement lawsuit for $1.5 billion, Warner Bros. Discovery sues Midjourney for copyright infringement (2025); Disney and Universal v. Midjourney, Inc. (commenced June 2025); Warner Bros. Discovery v. Midjourney (commenced 4 September 2025); Disney, Universal and Warner Bros. Discovery v. MiniMax (commenced 16 September 2025).

New York Attorney General Letitia James commenced proceedings on 21 April 2026 against Coinbase Financial Markets, Inc. and Gemini Titan, LLC over their prediction market products, seeking illegal-profit disgorgement, restitution, civil fines equal to three times the operators’ profits and reported minimum damages of $2.2bn from Coinbase and $1.2bn from Gemini.¹

The complaints were filed fifteen days after the Third Circuit held in KalshiEX LLC v Flaherty, No. 25-1922, that the Commodity Exchange Act likely preempts state gambling enforcement against sports-related event contracts traded on CFTC-licensed Designated Contract Markets.² The alignment of CFTC registration, state gambling theory and federal circuit geography now determines the litigation risk for every exchange offering event contracts.

The preemption architecture

Judge David Porter’s majority opinion in Kalshi, joined by Chief Judge Michael Chagares and affirming the District of New Jersey’s preliminary injunction against the New Jersey Division of Gaming Enforcement, rested on two bases. Field preemption followed from the CFTC’s exclusive jurisdiction over swaps, including sports-related event contracts, traded on DCMs. Conflict preemption followed from the statutory purpose of avoiding the state-by-state regulatory patchwork that the Commodity Exchange Act was designed to displace.³ Judge Jane Richards Roth’s dissent, treating Kalshi’s contracts as sports betting in structural fact, keeps the doctrinal question live for state regulators.

Gemini Titan received its DCM designation on 10 December 2025. Coinbase Derivatives, LLC, the group’s DCM affiliate acquired via FairX, has held DCM status since 23 November 2020. Coinbase Financial Markets, Inc., the entity James has sued, is registered as an FCM. On 2 April 2026, the CFTC commenced federal proceedings against Arizona, Connecticut and Illinois, asserting exclusive jurisdiction over event contracts traded on registered DCMs. On 8 April, a federal court in Arizona granted preliminary injunctive and temporary restraining relief on the same theory. By 21 April, the preemption argument had been accepted or endorsed by several federal courts, including the Third Circuit and, most recently, the District of Arizona.⁴

The Second Circuit question

New York sits in the Second Circuit, which has not ruled on CEA preemption in the event-contract context. The Third Circuit opinion is persuasive authority, not binding authority. Its reasoning nevertheless transports without much strain to event contracts traded on CFTC-licensed DCMs. The Second Circuit has previously recognised broad CEA preemption in the futures context, and section 2(a)(1)(A) of the Commodity Exchange Act contains the exclusive-jurisdiction language on which the Kalshi majority relied.⁵

The entity split matters to New York’s pleading design. Coinbase Financial Markets is an FCM. Coinbase Derivatives is the DCM and is not a named defendant. The preemption defence, which runs through DCM registration under section 2(a)(1)(A), therefore reaches the DCM directly but reaches Coinbase Financial Markets only to the extent that the event contracts at issue are allocated to, or transacted through, the DCM affiliate. Gemini Titan carries its DCM registration in its own name and has the cleaner preemption posture of the two defendants.

The litigation design tells us something about the target. New York is attempting to build a doctrinal record in a circuit not yet bound by Kalshi, while Kalshi itself, which James has not named, remains the structurally strongest preemption defendant.

Practitioner implications

For crypto-native exchanges offering event contracts, three questions now follow. First, whether product migration from FCM-facing to DCM-facing entity structures is required for full preemption coverage against state gambling enforcement. Second, whether the New York actions should be removed to federal court and consolidated, coordinated or otherwise sequenced with federal proceedings raising the same preemption issue. Third, whether state attorneys general will treat CFTC registration as a negotiating chip rather than a jurisdictional bar.

The structural issue is adjacent to the federalism problem Justice Alito identified in Murphy v National Collegiate Athletic Association: where Congress regulates directly and does so with sufficient clarity, preemption turns on statutory text rather than state gaming authorities’ characterisation of the product.⁶ The Third Circuit has resolved the question at preliminary-injunction stage for sports-related event contracts traded on CFTC-licensed DCMs. The Second Circuit, sooner rather than later, will be asked whether to follow. Whether it does, and on what reasoning, will shape the next generation of prediction-market litigation and the asset class it surrounds.

¹ Office of the NY Attorney General press release, Attorney General James Sues Coinbase and Gemini for Running Illegal Gambling Platforms in New York (21 April 2026). Reuters, CoinDesk, CNBC, American Banker and CBS New York (21 April 2026). The proceedings were commenced in New York state court in Manhattan.

² KalshiEX LLC v Flaherty, No. 25-1922 (3d Cir., decided 6 April 2026). Slip opinion via Justia. Skadden, Third Circuit Affirms Kalshi’s Preliminary Injunction (April 2026). Paul, Weiss, A Divided Third Circuit Holds That the CFTC Has Exclusive Jurisdiction Over Sports-Related Event Contracts (April 2026).

³ KalshiEX LLC v Flaherty, slip opinion. For earlier federal decisions considering the preemption theory, see the Kalshi majority’s citations to M.D. Tenn. (19 February 2026), N.D. Cal. (10 November 2025) and D.N.J. (28 April 2025).

⁴ Gemini Investor Relations release, Gemini Receives US License for Prediction Markets (10 December 2025). CFTC DCM Registry entries for Gemini Titan, LLC and Coinbase Derivatives, LLC (legacy LMX Labs / FairX, designated 23 November 2020). CFTC release, 2 April 2026, multi-state proceedings against Arizona, Connecticut and Illinois. District of Arizona, Preliminary Injunction and Temporary Restraining Order (8 April 2026), available via the CFTC case portal.

⁵ Commodity Exchange Act, section 2(a)(1)(A). Specific Second Circuit CEA preemption authority to be added by the author on publication.

⁶ Murphy v National Collegiate Athletic Association, 584 US 453 (2018). The Court struck down the Professional and Amateur Sports Protection Act on anti-commandeering grounds and discussed preemption principles in the course of doing so. Murphy does not itself resolve the CEA / event-contract preemption question.

EvenUp, a generative AI platform designed for personal injury firms, achieved a $2 billion valuation in 2025. This single metric captures a structural shift in American litigation. The plaintiff bar has spent billions on technology to aggregate settlement data, standardise demand packages and model carrier-specific risk tolerances. The defence side remains fragmented. Individual insurance carriers and their panel counsel operate within data silos that obscure broader patterns in judicial behaviour and plaintiff strategy.¹

This divergence is driven by asymmetric AI adoption and the ways both sides collect, aggregate and utilise litigation data.¹ The insurance industry has long attributed rising liability costs to “social inflation” and “nuclear verdicts” as external, uncontrollable phenomena driven by shifting societal norms and anti-corporate bias. Emerging evidence suggests a different explanation: much of what the market experiences as uncontrolled cost growth is the direct result of coordinated technological advantage on the plaintiff side.

Plaintiff firms have transitioned from isolated practices to a technologically integrated ecosystem. They leverage AI to standardise aggressive demand packages and use crowdsourced settlement data to identify carrier-specific leverage points. The defence side has not matched this coordination. Individual insurance carriers and their counsel operate in isolation, obscuring broader market patterns and failing to develop shared intelligence about plaintiff tactics or judicial outcomes.¹

This structural imbalance drives up litigation costs and sends inconsistent signals to the market. When defence outcomes vary widely across materially similar cases, the plaintiff bar identifies these variations as opportunities. High settlements (even outliers) become the new reference points for future negotiations. The market amplifies what should be treated as noise into trend.

The Plaintiff-Side Technological Vanguard: Coordination at Scale

The plaintiff bar’s adoption of artificial intelligence is rooted in economic incentive. Operating on a contingency fee basis, plaintiff firms maximise returns by adopting any technology that reduces manual labour, shortens case lifecycle and increases settlement values.^4,5 By 2024, approximately 30% of legal professionals reported using AI technology, a 19% increase from the previous year, with personal injury practices at the forefront of this adoption curve.⁵

Automated Demand Engineering

The demand letter, historically one of the most time-consuming tasks in personal injury practice, has become automated. Platforms such as EvenUp, Supio and Novo ingest thousands of pages of unstructured data (medical records, police reports, billing statements) and generate settlement-ready demand packages in minutes.^6,8,9 These tools extract the most salient facts, map injuries to specific ICD-10 codes, and highlight pain and suffering narratives that manual review may overlook.^8,10

EvenUp’s “Mirror Mode” allows firms to train AI to draft documents that match their distinctive tone and style. The firm maintains efficiency without sacrificing brand consistency.³ These systems proactively identify missing medical records or treatment gaps, allowing firms to maximise claim value before the initial demand. This level of preparation forces adjusters into a defensive posture, presented with data-rich demands that are difficult to challenge without equivalent technological support.¹³

Crowdsourcing Private Settlements: Neutralising Information Asymmetry

The most notable strategic shift in the plaintiff bar is the move away from negotiating cases in isolation.¹ Historically, the defence side held an information advantage. Insurers handled thousands of claims and understood the “going rate” for specific injuries in specific venues. Plaintiff-side AI firms have neutralised this advantage by convincing thousands of firms to share their actual, anonymised settlement data.³

This crowdsourced data allows AI platforms to analyse not just public jury verdicts (which represent a tiny fraction of civil cases) but the private settlements that constitute the vast majority of resolutions.^3,14 By synthesising this collective intelligence, plaintiff attorneys can identify venue-specific leverage points, judicial tendencies and carrier-specific settlement patterns.¹ They can determine which adjusters are most likely to settle early and at what values, allowing them to calibrate their demands accordingly.¹

This coordination enables a standardisation of aggressive positions across large inventories of cases. No potential value remains on the table. The scale of investment supporting this shift is substantial. EvenUp’s $2 billion valuation signals that the plaintiff bar has reached a tipping point in its ability to leverage data as a strategic asset.³

The Defence Intelligence Gap: Fragmentation and Operational Friction

The insurance defence industry stands in stark contrast to the coordinated ecosystem of the plaintiff bar. This fragmentation is both organisational and industry-wide, creating systemic weaknesses that prevent the defence from countering the plaintiff side’s data-driven strategies.¹

The Internal Silo Problem

Within a typical insurance organisation, data is siloed between departments and individual adjusters. A claims professional in one region may have little visibility into how similar cases are being resolved in another, or even how different adjusters within the same office are valuing identical injuries.¹ This lack of internal communication produces uncertain valuation anchors. Negotiators struggle to distinguish between a fair evolution in case value and inflated “drift” caused by aggressive plaintiff tactics.¹ When defence outcomes vary widely across materially similar cases, it creates inconsistent market signals.¹

The plaintiff bar, using its coordinated AI tools, identifies these variations as opportunities. High settlements quickly become new reference points for future negotiations.¹ Because the defence side lacks a shared view of its own portfolio’s value, it often fails to recognise when a specific strategy is being tested across multiple jurisdictions until it has already become established as a trend.¹

The Hourly Billing Barrier to AI Adoption

Defence law firms traditionally operate on an hourly billing model. This creates a misalignment of incentives. Time-saving technologies that would benefit the insurer by reducing legal spend may be perceived as a threat to the firm’s revenue.⁴ This friction leads to a cautious stance toward AI adoption.

Whilst plaintiff firms are integrating generative AI to summarise medical records and draft pleadings, many defence teams are still manually sifting through voluminous discovery materials.^4,15,16 The manual process is slower and more prone to human error. AI can identify pre-existing conditions or inconsistencies in a plaintiff’s medical history in seconds.^15,16

Deconstructing Social Inflation as a Negotiated Phenomenon

The insurance industry has long used the term “social inflation” to describe the rise in liability claims costs above general economic inflation.^18,19 This label encompasses several factors, including anti-corporate bias and the rollback of tort reform. The available evidence suggests that much of what is attributed to social inflation is instead a portfolio phenomenon driven by the intelligence gap.¹

Quantifying Loss Severity

Actuarial research by the Insurance Information Institute and the Casualty Actuarial Society indicates that “legal system abuse” (‘LSA’) contributed between $231.6 billion and $281.2 billion in increased liability insurance losses between 2014 and 2023.²⁰ This surge outpaces general economic inflation as measured by the CPI-U.²⁰

The data confirms that these losses are driven by claim severity rather than frequency.^20,21 The number of claims filed has remained relatively stable or declined in some segments. The average payout per claim has soared.^20,21 This increase in severity results from more effective anchoring strategies employed by the plaintiff bar, often supported by AI-generated demand packages that highlight the most extreme potential damages.^10,22

Methodological Concerns with the LSA Framework

The insurance industry has attributed escalating liability costs to legal system abuse, pointing to sophisticated plaintiff tactics, third-party litigation funding and technology-driven demand packages as culprits behind claim severity that outpaces economic inflation.²⁰

This explanation warrants scrutiny. The LSA framework suffers from procedural weaknesses beginning with definitional circularity. Unexplained loss increases are simply labelled as “abuse” without rigorous proof of causation. The research fails to control adequately for legitimate cost drivers: medical inflation (which has consistently outpaced CPI for decades), escalating vehicle repair costs for technology-laden modern automobiles, genuine increases in injury severity, and wage growth affecting lost earnings calculations. Triple-I is an industry advocacy organisation rather than an independent research body. Its framing of increased costs as “abuse” may reflect an ideological perspective rather than objective analysis.

An alternative explanation for rising insurance premiums challenges the legal system abuse narrative. Cost escalation may result instead from systemic inefficiencies and poor coordination within the defence bar and insurance industry itself. Whilst plaintiff attorneys have invested heavily in technology, data analytics and coordinated litigation strategies, the defence side has remained fragmented. Individual insurers and their counsel work in silos without sharing intelligence or best practices. This lack of strategic coordination means defence counsel repeatedly reinvent solutions for similar cases, fail to leverage collective data on settlement patterns, and miss opportunities for early intervention that would reduce claim severity.

Many insurers have been slow to adopt the technological tools now standard among plaintiff firms, relying instead on outdated case management systems and reactive rather than proactive defence strategies. The result is not legal system abuse but a predictable market response: when one side invests in efficiency and intelligence whilst the other does not, costs naturally rise. The premium increases may reflect the insurance industry’s own failure to modernise and coordinate its defence efforts, rather than any distortion of the legal system.

The Role of Third-Party Litigation Funding

The escalation of litigation costs is further fuelled by the growth of third-party litigation funding, a global industry that enables investors to finance lawsuits in exchange for a portion of the settlement.^18,23,24 TPLF provides the capital necessary for plaintiff firms to invest in the technology widening the intelligence gap.^1,24 Analysis by the National Insurance Crime Bureau suggests that 75% of assessed insurance companies have been directly targeted by litigation marketing campaigns backed by funders.²⁴

Some observers suggest this funding transforms individual personal injury claims into an asset class, incentivising longer litigation cycles and higher settlement demands as funders seek to maximise their return on investment.^18,23

Proponents of third-party litigation funding argue that it serves a vital access-to-justice function, particularly for claimants lacking the financial resources to pursue meritorious claims against well-funded defendants. Without such funding mechanisms, many legitimate claims would remain unpursued due to prohibitive litigation costs, effectively denying redress to those who have suffered genuine harm. TPLF helps level the playing field, enabling individuals and smaller entities to assert their legal rights against corporations and insurers with substantially greater financial capacity. From this perspective, funding democratises access to the legal system rather than distorting it, ensuring that the ability to seek justice is not determined solely by financial means.

Towards a Defence AI Ecosystem: Restoring the Strategic Balance

To counter the coordinated intelligence of the plaintiff bar, insurance organisations and defence firms must adopt AI-driven tools and foster industry-wide data sharing.¹ This transition requires moving beyond AI as a tool for document review. Integration into the core of negotiation and portfolio management is essential.¹

Predictive Modelling and Valuation Consistency

Emerging platforms designed for the defence side focus on restoring valuation anchors and providing defensible settlement recommendations.^13,25 Systems like SigmaSight provide AI-driven analytics that generate risk and valuation ranges grounded in economic data, venue trends and specific injury types.¹³ These tools allow claims professionals to negotiate with confidence. Every offer is supported with a structured Offer Package that clearly explains the rationale to mediators and opposing counsel.¹³

AI for predictive settlement modelling enables insurers to estimate settlement ranges and probabilities more accurately, tightening reserve bands and reducing late-stage financial surprises.²⁵ By identifying outlier cases early, these systems allow defence teams to anticipate plaintiff strategies before they result in large verdicts.¹³

Counsel Collaboration and Pattern Recognition

AI can serve as a bridge between insurance carriers and their defence counsel. By sharing AI-backed insights, claims professionals can ensure their attorneys are aligned on valuation and strategy from the outset.¹³ This collaboration allows the defence to develop more effective responses against specifically aggressive plaintiff firms.²⁶ CLARA Analytics provides detailed scorecards for defence counsel, benchmarking them against industry peers and analysing their past performance against specific plaintiff attorneys.²⁶

This shift addresses the pattern recognition deficit on the defence side.¹ By aggregating data across portfolios, AI can identify when a particular plaintiff firm is deploying a new tactical narrative (such as “junk science” theories or novel pain and suffering arguments), allowing the defence to develop a coordinated response rather than reacting to each case in isolation.¹

Ethical Guardrails and the Judicial Response to AI

The rapid integration of AI into legal practice has created new ethical risks, with AI hallucinations emerging as the most visible: generative tools fabricate non-existent case citations.^28,29,30 High-profile incidents like Mata v. Avianca and Johnson v. Dunn resulted in substantial judicial sanctions and a sharp rebuke of attorneys who failed to exercise appropriate oversight.^29,30

Human-in-the-Loop Oversight

These judicial developments highlight a requirement for AI adoption on both sides: the human-in-the-loop model.^27,31 Ethical guidelines from the American Bar Association and state bars emphasise that AI tools must remain under human authority, oversight, and control.^5,32 For the defence side, where professional credibility and strict confidentiality are paramount, this human oversight is non-negotiable.^16,17,27 AI should perform data-intensive tasks of extraction and summarisation. Final legal judgment and strategic decisions must rest with the attorney.^4,16

Privilege and the Digital Stranger

A concern for insurers is the potential waiver of attorney-client privilege when using public AI platforms.^17,33 Public models often retain uploaded data for training purposes, potentially exposing sensitive litigation strategies to a “digital stranger” outside the protected relationship.¹⁷ To mitigate this risk, defence organisations are moving toward private or firm-built AI systems that operate under strict contractual relationships ensuring data segregation and confidentiality.^17,34 A private AI system functions more like an agent of the firm (an expert witness or e-discovery consultant) rather than an unregulated third party.¹⁷

Regulatory Evolution and the 2025 Roadmap

As the impact of AI on litigation becomes clearer, regulators are taking proactive steps to ensure market stability and consumer protection.^35,36 The National Association of Insurance Commissioners (‘NAIC’) has identified AI governance as a top priority in its 2025 roadmap, “Securing Tomorrow: Advancing State-Based Regulation”.³⁵

NAIC Initiatives

The NAIC’s 2025 strategy focuses on enhancing financial governance and modernising risk-based capital frameworks to address the demands of the evolving market.³⁵ Specifically, the NAIC is working on frameworks for third-party data and predictive models, emphasising outcome transparency.³⁶ States such as New York, Colorado and Connecticut are expected to lead in “AI outcomes testing,” requiring insurers to demonstrate that their use of AI does not result in unfair bias or discriminatory practices.³⁶

The NAIC is expected to introduce a new privacy protections model law by late 2025, addressing the risks associated with the massive amounts of data (estimated to reach 181 zettabytes annually) being generated and consumed by AI systems.^36,38 This regulatory pressure will force insurers to be more deliberate and transparent in their use of AI for claims and litigation management.^36,39

Re-Examining Collective Intelligence: The Industry Data Mandate

The final requirement for restoring balance is the need for industry-wide coordination on the defence side.¹ If the plaintiff bar is benefiting from collective intelligence through shared settlement repositories, the defence industry must confront whether its traditional norms of isolation are inadvertently causing financial harm.¹

Other sectors of the insurance industry already rely on pooled data to manage risk. The National Insurance Crime Bureau operates shared fraud detection systems. Catastrophic modelling relies on shared databases.¹ Litigation negotiation remains a notable exception, where each carrier negotiates without the context of the broader market.¹

Establishing contributory databases or shared benchmarking would allow the defence side to achieve parity with the tools the plaintiff bar is already using.¹ This does not require perfect data, but rather a commitment to coherence and proportionality.¹ By creating a unified front of data, insurers can prevent the anchoring of settlements at unsustainable levels and ensure that individual outliers do not recalibrate the entire market’s expectations.¹

The ultimate goal of adopting AI and data-sharing is to strengthen, rather than replace, professional judgment.^1,13,40 AI functions as a navigational tool for adjusters, providing structured insights that help them analyse complex files more effectively.^25,26 By automating the repetitive aspects of case analysis (such as medical billing review and chronology generation), AI frees up human professionals to focus on high-value tasks of strategy, advocacy, and negotiation.^13,41,42

Conclusion: Restoring Equilibrium in the AI Era

The growing intelligence gap in litigation is not an inevitable byproduct of technological advancement. It is the result of a coordinated strategic shift by the plaintiff bar that has yet to be fully met by the defence.¹ Social inflation is a real and measurable force. Its impact, however, is amplified by the defence side’s internal fragmentation and lack of data-driven negotiation tactics.¹

Restoring equilibrium requires a dual response. The defence must integrate AI-driven tools into the claims and litigation workflow to improve valuation consistency, reduce operational costs, and sharpen defence arguments.^13,15 Beyond the firm level, the insurance industry must abandon the silo mentality and develop mechanisms for collective intelligence and shared benchmarking to counter the plaintiff bar’s coordinated advantage.¹

By adopting a proactive approach to AI and data sharing, the defence side can transition from a reactive posture (often blamed on external societal shifts) to one of professional control and financial stability.^1,13 The future of casualty litigation will be defined by those who can most effectively synthesise the vast oceans of available data into actionable, defensible intelligence.^26,38 Restoring balance at the negotiating table is not merely a technological challenge. It is a structural mandate for the preservation of a fair and sustainable legal system.^1,40

References

¹ The Defense Intelligence Gap: Why Litigation Portfolio Values Are Accelerating in an AI Era, https://riskandinsurance.com/the-defense-intelligence-gap-why-litigation-portfolio-values-are-accelerating-in-an-ai-era/

² Turning the tide on social inflation and rising liability insurance costs—Lockton, https://global.lockton.com/us/en/news-insights/turning-the-tide-on-social-inflation-and-rising-liability-insurance-costs

³ From Startup To $2 Billion: EvenUp Is Transforming Personal Injury, https://www.techlawcrossroads.com/2025/10/from-startup-to-2-billion-evenup-is-transforming-personal-injury-practice/

⁴ How Plaintiff Law Firms Are Outpacing Defense Firms in AI Adoption—And What to Do About It, Wisedocs, https://www.wisedocs.ai/blogs/how-plaintiff-law-firms-are-outpacing-defense-firms-in-ai-adoption

⁵ Using emerging AI tech to prompt your future, Advocate Magazine, https://www.advocatemagazine.com/article/2025-august/using-emerging-ai-tech-to-prompt-your-future

⁶ Eve | Legal AI for plaintiff firms, https://www.eve.legal/

⁷ The Top 5 Benefits of AI for Personal Injury Law Firms, Eve Legal, https://www.eve.legal/blogs/the-top-5-benefits-of-ai-for-personal-injury-law-firms

⁸ AI Demand Letter Drafting Platform for Personal Injury Firms, Supio’s AI, https://www.supio.com/products/ai-demand-letter

⁹ AI Demand Letter Tools and Strategies for Personal Injury Firms, CasePeer, https://www.casepeer.com/blog/ai-demand-letter/

¹⁰ EvenUp | Personal Injury AI — From Intake to Resolution, https://www.evenuplaw.com/

¹¹ Negotiation Preparation™ | AI-led Settlement Negotiations Platform, EvenUp, https://www.evenuplaw.com/products/negotiation-preparation

¹² Improve Demand Letter Effectiveness with AI-Powered Insights, precedent.com, https://precedent.com/the-rise-of-pi-automation-2025-outlook/

¹³ Claim Professionals, SigmaSight, https://www.sigmasight.ai/claim-professionals

¹⁴ Policyholders Are Not to Blame for “Social Inflation”, https://www.pillsburylaw.com/en/news-and-insights/policyholders-not-to-blame-for-social-inflation.html

¹⁵ AI for Insurance Carriers | CaseMark—Pre-Approved AI for Defense Teams, https://www.casemark.com/solutions/insurance-carriers

¹⁶ Enhancing Legal Defense Strategies with AI Medical Summaries, Wisedocs, https://www.wisedocs.ai/blogs/enhancing-legal-defense-strategies-with-ai-medical-summaries

¹⁷ Digital Strangers in Litigation: Does Sharing with AI Breach Privilege?, TransPerfect Legal, https://www.jdsupra.com/legalnews/digital-strangers-in-litigation-does-6854504/

¹⁸ Insurance Topics | Social Inflation, NAIC, https://content.naic.org/insurance-topics/social-inflation

¹⁹ Social Inflation: Navigating the evolving claims landscape, The Geneva Association, https://www.genevaassociation.org/sites/default/files/social_inflation_brief_web.pdf

²⁰ Legal System Abuse, Not Just Economic Inflation, Drives Liability Insurance Losses by More than $230 Billion Over Past 10 Years, Insurance Information Institute, https://www.iii.org/press-release/legal-system-abuse-not-just-economic-inflation-drives-liability-insurance-losses-by-more-than-230-billion-over-past-10-years-new-triple-i-casualty-actuary-society-analysis-shows-103025

²¹ What Is the Evidence for Social Inflation?, RAND, https://www.rand.org/pubs/research_reports/RRA2645-1.html

²² Three Key Drivers of Social Inflation, AJG United States, https://www.ajg.com/news-and-insights/features/social-inflation-nuclear-verdicts-drivers/

²³ 4 Social Inflation Drivers Contributing to Rising Claim Costs, Travelers Insurance, https://www.travelers.com/resources/business-topics/insuring/4-factors-causing-social-inflation

²⁴ NICB Warns Insurance Industry of Fraud Facilitated by Third-Party Litigation Funding, https://www.nicb.org/news/news-releases/nicb-warns-insurance-industry-fraud-facilitated-third-party-litigation-funding

²⁵ Negotiated Settlement Impact AI Agent for Claims Economics in Insurance, insurnest, https://insurnest.com/agent-details/insurance/claims-economics/negotiated-settlement-impact-ai-agent-for-claims-economics-in-insurance/

²⁶ Litigation Analytics – Turning Data into a Competitive Advantage for Casualty Insurers, Genre, https://www.genre.com/us/knowledge/publications/2025/august/litigation-analytics-turning-data-into-a-competitive-advantage-for-casualty-insurers-en

²⁷ AI, Trust, and the Human Touch: Wisedocs’ Webinar with CLM, https://www.wisedocs.ai/blogs/ai-trust-and-the-human-touch-clm-webinar

²⁸ Artificial Intelligence or innocent ignorance? Hard lessons yield best practices, Clark Hill, https://www.clarkhill.com/news-events/news/artificial-intelligence-or-innocent-ignorance-hard-lessons-yield-best-practices/

²⁹ Will AI Render Lawyers Obsolete?, New York State Bar Association, https://nysba.org/will-ai-render-lawyers-obsolete/

³⁰ Federal Court Turns Up the Heat on Attorneys Using ChatGPT for Research, https://www.esquiresolutions.com/federal-court-turns-up-the-heat-on-attorneys-using-chatgpt-for-research/

³¹ How AI Is Reshaping Litigation: Insights from DRI’s Senior Living Seminar, Excelas, https://excelas1.com/how-ai-is-reshaping-litigation-insights-from-dris-senior-living-seminar/

³² Law-Related Artificial Intelligence: Ethics Issues, McGuireWoods, https://media.mcguirewoods.com/publications/Ethics-Programs/88951932.pdf

³³ Your AI-Powered Practice: Trial Preparation, Ethics, and the Future, Sam Aguiar Injury Lawyers, https://aguiarinjurylawyers.com/your-ai-powered-practice-trial-preparation-ethics-and-the-future/

³⁴ The AI-Enabled Insurance Defense Firm: A Strategic Roadmap for an Era of Disruption, https://www.performlaw.com/law-firm-best-practices-blog/the-ai-enabled-insurance-defense-firm-a-strategic-roadmap-for-an-era-of-disruption

³⁵ NAIC Announces 2025 Initiatives, https://content.naic.org/article/naic-announces-2025-initiatives

³⁶ 2025 Insurance Regulatory Outlook, Deloitte US, https://www.deloitte.com/us/en/services/consulting/articles/insurance-regulatory-outlook.html

³⁷ Commissioner Lara Announces Major Reforms to the Intervenor Process, California Department of Insurance, https://www.insurance.ca.gov/0400-news/0100-press-releases/2025/release059-2025.cfm

³⁸ The Two Faces of AI/Articles/CLM Magazine, Claims and Litigation Management Alliance, https://www.theclm.org/Magazine/articles/the-two-faces-of-ai/3218

³⁹ Overview | Data Protection & AI, Lewis Rice, https://www.lewisrice.com/data-protection-ai

⁴⁰ DRI The Voice, August 2025, https://www.dri.org/newsletters/the-voice/2025/august

⁴¹ AI-Powered Defense (Part 2): Better Outcomes, Less Effort, More Profit, DRI Communities, https://community.dri.org/events/event-description?CalendarEventKey=98853280-3b56-4cb7-8de4-01996551889b&Home=%2Fevents%2Fcalendar

⁴² The Rise of AI/Articles/CLM Magazine, Claims and Litigation Management Alliance, https://www.theclm.org/Magazine/articles/the-rise-of-ai-in-workers-comp-claims/2873

A federal judge has dismissed the federal securities claims arising from Caitlyn Jenner’s JENNER memecoin. Public reporting on the order indicates that the complaint failed at the ‘common enterprise’ stage of the Howey analysis. The ruling matters less for Jenner than for the pleading architecture of memecoin class actions. The court’s reasoning and what it declined to decide, will shape the next wave of memecoin litigation.

The JENNER memecoin was launched on Solana on 26 May 2024 and on Ethereum in late May and early June of that year. Promotion ran through Caitlyn Jenner’s social channels, including posts on X using AI-generated imagery that suggested profit potential. The Solana token peaked at a market capitalisation of around $43 million and collapsed within weeks. The Ethereum version repeated the pattern on a smaller scale. Purchasers brought a putative class action in November 2024, pleading that the offerings were unregistered securities under Sections 5 and 12(a)(1) of the Securities Act 1933 and seeking rescission and damages. The Second Amended Complaint was the pleading before Blumenfeld J on the motion to dismiss. The action sits in a first wave of federal securities class actions against celebrity-promoted memecoin launches, alongside the pending Hawk Tuah action in the Eastern District of New York.

The ruling

In Naeem Azad et al v Caitlyn Jenner et al,¹ Judge Stanley Blumenfeld Jr of the US District Court for the Central District of California entered an order granting in part the defendants’ motion to dismiss the Second Amended Complaint and a separate final judgment on 16 April 2026. Federal securities claims under Sections 5(a), 5(c) and 12(a)(1) of the Securities Act were dismissed with prejudice as to the lead plaintiff, Lee Greenfield. The court declined to exercise supplemental jurisdiction over the common-law fraud and quasi-contract claims, which were dismissed without prejudice. The case was terminated the same day.

The class action had been commenced in November 2024 following the launch of JENNER on Solana on 26 May 2024 and the subsequent Ethereum launch in late May and early June 2024, followed by a collapse in market price. The plaintiffs relied on Jenner’s promotional activity across X, a 3% transaction tax on the Ethereum-based token and promised buyback and exchange-listing efforts.

The doctrinal move

The court’s reasoning is disciplined. Public reporting on the order indicates that Blumenfeld held the Second Amended Complaint failed to plead horizontal commonality under SEC v W J Howey Co.² The pleaded features, celebrity promotion, AI-generated imagery, a 3% transaction tax and buyback and listing promises, did not show investors had agreed to split profits. Because common enterprise was not made out, the court did not need to consider whether profits were expected from the efforts of others.

The ruling turns on Howey, not on the SEC staff statement of 27 February 2025.³ That statement is expressly non-binding by its own terms, and Commissioner Crenshaw publicly contested its analytical basis the same day.⁴ What Blumenfeld has done is apply the doctrine and the order will travel beyond the case for that reason.

The contrast with the Hawk Tuah litigation is instructive.⁵ That action, pending in the Eastern District of New York, alleges a pre-sale, airdrops and whitelists, a 17% strategic allocation routed through an offshore structure and the use of pooled funds for development and marketing. Those allegations present a more elaborate distribution structure than the allegations in Azad v Jenner and will be analysed on different facts. Common enterprise may be pleaded differently where purchasers’ contributions are said to have been pooled for a common purpose beyond the act of purchase.

The comparators and what survives

Under Regulation (EU) 2023/1114 (MiCA), crypto-assets without an identifiable issuer fall outside the Title II white paper obligations.⁶ Crypto-asset service providers handling distribution remain regulated under Title V. MiCA supplies a taxonomy of asset-referenced tokens, e-money tokens and other crypto-assets. Which category a given memecoin falls into turns on the token’s features and its manner of offer and sale, not on branding.

In the UK, financial promotions for qualifying cryptoassets are already within scope of the financial promotion regime.⁷ The wider UK cryptoasset regulatory regime is expected to activate on 25 October 2027, with the Consumer Duty extending to authorised cryptoasset firms.⁸ The FCA’s January 2026 Consumer Duty consultation and associated materials focus on clearer risk communication and appropriateness testing for the sector as a whole.

The narrower proposition is the right one. Azad v Jenner makes common enterprise a more visible pleading vulnerability in celebrity memecoin cases. It does not foreclose different outcomes on different records and differently structured token sales will be analysed on their own facts. What survives across the class action docket is state consumer protection, common law fraud and untested arguments over arbitration clauses embedded in custody and wallet terms. Memecoin litigation will migrate rather than disappear.

Code on Trial covers AI, crypto and digital disputes. Subscribe free for weekly analysis at www.codeontrial.ai

Footnotes

1. Naeem Azad et al v Caitlyn Jenner et al, Case No. 2:24-cv-09768-SB-JC (CD Cal), Final Judgment, 16 April 2026 (Blumenfeld J). The final judgment records that a same-day order was entered dismissing the federal securities claims and declining supplemental jurisdiction over the common-law fraud and quasi-contract claims.

2. SEC v W J Howey Co, 328 US 293 (1946).

3. SEC Division of Corporation Finance, Staff Statement on Meme Coins (27 February 2025). The statement records that its view is not dispositive and that determinations depend on the specific facts and the manner of offer and sale.

4. Commissioner Caroline A Crenshaw, Response to Staff Statement on Meme Coins: What Does it Meme? (27 February 2025).

5. See class action complaint against the promoters of $HAWK, pending in the Eastern District of New York (commenced December 2024). Pleading detail drawn from public case materials and firm disclosures.

6. Regulation (EU) 2023/1114 of the European Parliament and of the Council of 31 May 2023 on markets in crypto-assets; ESMA Questions and Answers on MiCA (February 2026).

7. FCA Policy Statement PS23/6 and subsequent supervisory materials on the financial promotion regime for qualifying cryptoassets.

8. Financial Conduct Authority, A new regime for cryptoasset regulation; statutory instrument laid before Parliament on 4 February 2026; regime expected to activate on 25 October 2027. FCA consultation papers on the Consumer Duty for cryptoasset firms published in January 2026.

Tesla’s latest Autopilot loss exposes more than the scale of the verdict. It lays bare a doctrinal problem that courts are only beginning to confront. Product liability law was built for fixed products, not software-driven systems that update themselves after sale. The Benavides verdict did not resolve that tension, but it makes it impossible to ignore.

In brief

• Benavides is not just a large verdict. It is a warning signal for AI-era product litigation.

• Product liability doctrine assumes a designed defect in a comparatively stable product.

• Driver-assistance systems complicate that model because behaviour emerges from updates, sensor inputs and machine-learning architectures.

• The EU has started to adapt legislatively. US courts are still building doctrine case by case.

The Benavides Verdict

In Benavides v. Tesla (No. 1:21-cv-21940-BB, S.D. Fla.), a Miami jury found Tesla 33% at fault for a fatal 2019 Model S crash, awarding $42.57 million in compensatory damages on that basis and $200 million in punitive damages, for a total verdict of $242.57 million.¹ On 19 February 2026, Judge Beth Bloom denied Tesla’s post-trial motions, upholding the jury’s findings on defective design and failure to warn and rejecting Tesla’s renewed motion for judgment as a matter of law. An appeal is expected.

Electrek, an electric vehicle trade publication, estimated on 16 April 2026 that Tesla’s overall litigation and regulatory exposure across 21 active fronts could reach $14.5 billion.²

The outcomes elsewhere are not uniform. Tesla won Autopilot trials in Los Angeles in April 2023 and Riverside in October 2023.³ By the time Benavides went to trial, Tesla had already resolved or avoided trial in several similar Autopilot cases. The pattern is mixed, but the direction of the verdicts reaching juries is harder for the defence to dismiss than it was three years ago.

The Doctrinal Problem

The underlying doctrinal problem is real regardless of individual outcomes. Product liability doctrine assumes a manufacturer designs a product, the product has a defect, and the defect causes harm.⁴ Driver-assistance systems that update themselves over the air, whose behaviour derives from neural network weightings no engineer specifically authored and whose failures involve real-time interactions between software, sensor data and a human driver do not sit neatly within that framework. Courts are applying design defect, failure to warn and negligence theories because those are the tools available. Whether those tools are adequate for software-driven systems that change after sale is the question that Benavides forces but does not resolve.

The National Highways Traffic Safety Administration’s Standing General Order 2021-01 requires crash reporting for vehicles operating with Level 2 and above automated driving systems, but the regulatory framework treats Level 2 as a driver-support feature where the human driver must remain fully and continuously engaged.⁵ That classification matters for the doctrinal analysis. The systems currently in litigation are not legally autonomous. They are assistive features marketed in terms that invite reliance and then defended on the basis that the driver should have been paying attention. That tension sits at the centre of the product liability question.

The EU Comparator

The EU has moved further than current US case law, but not by generally reversing the burden of proof. Directive (EU) 2024/2853 extends the product liability regime to software-driven products and recognises that manufacturers may remain liable for defectiveness arising after sale from software or related services within their control, including updates, upgrades and certain machine-learning behaviour.⁶ It also introduces disclosure obligations and rebuttable presumptions where technical or scientific complexity makes proof unusually difficult. But the claimant still bears the primary burden of proving defectiveness, damage and causation. Only a substantial modification is treated as creating a new product; ordinary updates do not restart the limitation framework.

The Trajectory

US courts have no equivalent statutory regime, which is why the doctrine is being built case by case, with inconsistent results. The question is whether aggregate exposure on the scale now emerging will accelerate legislative intervention or whether courts will continue adapting traditional tort categories incrementally.

The doctrinal gap exposed by Benavides is not unique to Tesla or to driver-assistance systems. Any product whose behaviour is shaped by post-sale software updates and machine-learning processes will encounter the same analytical mismatch. The scale of the Benavides verdict and the size of the broader Tesla docket may make autonomous vehicles the arena in which this problem is resolved first, but the precedent will travel. If courts cannot adapt, the pressure will shift to legislatures.

References

1. Benavides v. Tesla, Inc., No. 1:21-cv-21940-BB (S.D. Fla.). Verdict of $242.57 million ($42.57 million compensatory, $200 million punitive); post-trial motions denied by Judge Beth Bloom, 19 February 2026.

2. Electrek, ‘Tesla Facing Up to $14.5 Billion in Lawsuits’, 16 April 2026.

3. Reuters reported Tesla victories in Autopilot trials in Los Angeles (April 2023) and Riverside (October 2023), alongside settlements and dismissals in other Autopilot-related cases.

4. Restatement (Third) of Torts: Products Liability (1998), § 2.

5. National Highway Traffic Safety Administration, Standing General Order 2021-01, Third Amendment (24 April 2025), requiring reporting of crashes involving Level 2+ automated driving systems.

6. Directive (EU) 2024/2853 of the European Parliament and of the Council of 23 October 2024 on liability for defective products, Arts 9–10.

The public conversation about artificial intelligence has focused heavily on hallucinations, on systems fabricating case citations or inventing facts. That problem is real. But it is not the most consequential form of AI failure now reaching the courts. The more pervasive category is predictive error. Systems that do not invent information, but that get the answer wrong. An algorithm that denies a Medicare patient post-acute care based on a recovery timeline disconnected from clinical reality. A fraud-detection system that accuses 40,000 people of welfare fraud with a 93% false positive rate. A hiring tool that tells a deaf applicant to practise active listening.

These are not glitches, they are the products of systems designed to apply statistical averages to individual circumstances and the litigation they have generated is redefining how courts allocate responsibility for automated decisions. This article examines predictive failures across healthcare, insurance, employment, public benefits, criminal justice, real estate and securities markets, identifies the legal doctrines emerging from the resulting litigation and assesses the accountability frameworks that are beginning to take shape.

Healthcare and Insurance

The nH Predict Algorithm and Medicare Advantage

The litigation involving UnitedHealthcare and its nH Predict algorithm is the most prominent example of a predictive system producing systematically incorrect outcomes in healthcare. nH Predict was developed to forecast how long patients would require care in skilled nursing facilities or post-acute settings. The tool generated recovery timelines that were then used as the basis for coverage determinations.^1,2

The problem was structural. nH Predict was calibrated to predict average length of stay, but it was applied as a functional ceiling on coverage. Internal investigations found that while the tool reduced average stays by 15% to 25%, it did so at the expense of accuracy.² The plaintiffs allege that over 90% of patient claim denials generated by the algorithm were reversed when challenged through internal appeals or federal Administrative Law proceedings, an allegation the court treated as sufficient at the pleading stage and one UnitedHealth disputes.² A reversal rate that high does not indicate a system identifying unnecessary care. It indicates a system systematically under-predicting the care required by legitimate patients.

In Estate of Gene B. Lokken et al. v. UnitedHealth Group, Inc., decided by Judge John R. Tunheim in the District of Minnesota on 13 February 2025, the court allowed a putative class action to proceed by focusing on the insurer’s contractual representations.³ If an insurer represents to policyholders that decisions will be based on clinical judgment but instead relies on a flawed algorithm, the plaintiffs can sue for breach of contract and breach of the implied covenant of good faith. The ruling separates the coverage denial from the use of the inaccurate tool, creating a distinct pathway for algorithmic accountability in healthcare.

Range Compression in Personal Injury Claims

In casualty and liability insurance, the shift from human-led negotiation to software-driven settlement recommendations has produced what claimants describe as range compression. Systems used by Allstate and Progressive categorise claims, flag fraud suspicion, and suggest settlement ranges based on standardised inputs such as diagnosis and procedure codes.¹ These systems often fail to account for variables that resist quantification. Pain intensity, long-term mobility impairment and caregiving burdens do not fit into structured data categories and algorithms that omit them will systematically undervalue the claims they assess.

In 2024, Progressive agreed to a $48 million settlement in New York to resolve a class action alleging that third-party software adjustments (Mitchell International’s WorkCenter Total Loss) had systematically undervalued total loss claims for approximately 93,000 policyholders.¹ In 2010, Allstate paid $10 million to 45 states following a multi-state NAIC examination of its use of the Colossus bodily injury claim software.¹ In both cases, the algorithm was optimised to reduce average payouts rather than to produce accurate valuations of individual losses. When the optimisation target is a business metric rather than an accuracy metric, undervaluation is not a bug. It is the system working as designed.

Employment and Algorithmic Agency

Mobley v. Workday

Mobley v. Workday, Inc. has become the leading case on liability for automated recruitment. The plaintiff, a 40-year-old black job seeker with a disability, alleged that Workday’s AI-driven applicant screening system systematically rejected his applications for nearly 100 positions.^4,5

In July 2024, the court allowed discrimination claims to proceed against Workday on an agency theory, while dismissing the separate employment-agency theory.⁶ The reasoning on the surviving claim was that the software did more than execute employer-defined criteria. It used its own AI models to evaluate, score and rank candidates, making it an active participant in the hiring decision rather than a passive tool.^6,7 If the agency theory succeeds at trial, it will prevent companies from outsourcing compliance obligations to technology vendors by holding both the employer and the vendor accountable for discriminatory outcomes.

Proxy Variables and Contextual Failure

Predictive hiring tools reproduce bias because they learn from historical data. If a company’s historical top performers have been predominantly white men from specific universities, an AI system will learn to rank candidates with those profiles higher, regardless of whether other candidates are objectively more qualified.^8,9 A 2024 University of Washington study tested 120 first names across three large language models against over 500 real job listings. The models preferred white-associated names 85% of the time and black-associated names 9% of the time.^4,8

In March 2025, the ACLU of Colorado filed a complaint against Intuit and HireVue regarding an AI video interview tool that gave a deaf employee feedback that she needed to practise active listening.^7,8 The system had misidentified her lack of auditory response as a lack of professional engagement. This was not a hallucination. It was a model that could not account for the context of disability and it illustrates the limits of systems that reduce human behaviour to statistical patterns.

Separately, Kistler et al. v. Eightfold AI Inc., filed in Contra Costa County Superior Court in January 2026 and removed to Federal court on 2 March 2026, raises an issue under the Fair Credit Reporting Act.⁶ The plaintiffs argue that AI-generated match scores, which rank candidates from 0 to 5 using social media and internet activity data, function as consumer reports. If that argument succeeds, vendors would be required to provide candidates with the right to access their scores and dispute inaccuracies, a requirement that most proprietary screening systems are not currently designed to meet.

Public Benefits and Automated Stategraft

When governments automate the adjudication of public benefits, the consequences of predictive error are borne by populations with the fewest resources to challenge them. Two cases illustrate the pattern.

Robodebt

Australia’s Robodebt scheme operated from 2015 to 2019. It was designed to identify welfare overpayments by comparing social security records with Australian Taxation Office data.^10,11 The algorithm divided annual income into fortnightly increments and compared them against reported fortnightly earnings. The method assumed income was earned evenly across the year. For anyone with variable or seasonal earnings, the assumption was wrong.

A student who earned nothing for ten months but worked full-time for two months would be flagged as having exceeded the income threshold for every fortnight of the year under the averaged model.¹¹ The mathematical error is explained by Jensen’s Inequality. Benefit eligibility is not a linear function of income. Averaging the input before applying the eligibility rules produces a different result than applying the rules to the actual data. The system generated more than 400,000 false debt notices.^10,13 The government settled the resulting class action for A$1.2 billion in 2020, though the total financial relief, including refunds and debt waivers, exceeded that figure.^10,13 The court ruled the scheme unlawful, in part because it shifted the burden of proof onto citizens to disprove debts generated by a flawed prediction.^12,14

Michigan MiDAS

Michigan’s Integrated Data Automated System was implemented to detect unemployment insurance fraud. Between 2013 and 2015, the system issued over 60,000 fraud determinations, wrongly accusing approximately 40,000 residents. Subsequent audits found a false positive rate of 93%.^15,16,17 The system was over-calibrated to treat any discrepancy between state and federal records as intentional misrepresentation. A typo in an employer’s name was sufficient to trigger a fraud determination.^15,17

Michigan levied a 400% penalty on alleged fraud, the highest in the nation.^17,18 Because the system also automated notifications, often sending them to out-of-date addresses, many victims were unaware of the accusation until their bank accounts were garnished.¹⁷ The litigation in Bauserman v. Unemployment Insurance Agency forced the state to acknowledge violations of procedural due process. The Michigan Supreme Court ruled that victims could recover monetary damages for constitutional-tort claims, and a $20 million settlement was approved by the Michigan Court of Claims in January 2024.^15,16

Criminal Justice and Predictive Policing

COMPAS and LSI-R

Risk assessment algorithms such as COMPAS and the LSI-R inform sentencing and parole decisions by predicting recidivism. These tools rely on static factors including education, family history and employment, variables that frequently serve as proxies for race and socioeconomic status.²⁰ The Department of Justice’s Criminal Division has described such assessments as dangerous because they penalise individuals for their social environment rather than their conduct.²⁰

ProPublica’s 2016 analysis found that black defendants who did not go on to reoffend were falsely flagged as high risk at a rate of 45%, compared with 23% for white defendants in equivalent circumstances. The disparity is roughly two to one.²⁰ White defendants were correspondingly more likely to be incorrectly classified as low risk. The model optimises for overall accuracy without ensuring equitable error rates across demographic groups.^7,21

Transparency Litigation

In EPIC v. DOJ, the Electronic Privacy Information Center sued for government records on risk assessments and predictive policing.²⁰ The case revealed a previously undisclosed DOJ report to the White House acknowledging the potential for disparate impacts and the erosion of consistent sentencing, even as the government continued to promote these tools.²⁰

In Chicago, journalists sued to obtain information about the city’s Strategic Subject List, an algorithmic tool that ranked individuals based on their likelihood of involvement in a shooting.²² The list was criticised for targeting individuals based on social networks and prior arrests for non-violent offences, leading to repeated police contact with people who had never been convicted of a serious crime.^22,23 The system identified a statistical correlation between an individual and a criminal network, but treated correlation as a basis for intervention, with the attendant infringement of civil liberties.^23,24

Real Estate, Credit Scoring and the Devaluation of Equity

Automated Valuation Models

AVMs are designed to provide objective property estimates for mortgage lending. Research by the Urban Institute found that these models systematically produce larger errors for black homeowners.²⁵ In a study of Atlanta and Memphis, AVMs generated valuation errors 3.4 percentage points higher for black homeowners than for white homeowners, and undervalued black-owned properties by an average of 5% compared to comparable white-owned homes.²⁵

The undervaluation reflects the model’s reliance on historical sales data from neighbourhoods shaped by segregation. If a neighbourhood has been historically devalued through redlining, the algorithm treats that devaluation as a market signal and incorporates it into future predictions.^25,26 The bias is not introduced by the model. It is inherited from the data and perpetuated through automation.

Algorithmic Credit Scoring

Financial institutions are increasingly using machine learning to analyse alternative data, including social media activity, browsing habits and device choices, to predict creditworthiness.²⁹ The intention is to serve the ‘credit invisible’, but the method introduces proxy variables that correlate with race and class. A model might predict that an individual who uses a particular device or visits certain websites is less likely to repay a loan, regardless of their actual financial history.²⁹

A 2022 Bloomberg investigation found that Wells Fargo’s mortgage algorithm approved only 47% of black refinancing applicants, compared with 72% of white applicants.²⁹ The failure was one of representation bias. The model was trained on data that did not adequately represent the financial behaviours of minority communities, producing skewed results that excluded high-earning minority borrowers from favourable loan terms.^29,30

Securities Fraud and AI-Washing

A growing area of litigation involves companies making misleading claims about the capabilities of their predictive analytics to attract investment. When the systems fail to deliver, or when it emerges that the claimed AI was largely non-functional, the result is securities fraud exposure.

In Helo v. Sema4 Holdings Corp, a securities fraud class action filed in Connecticut Federal Court, the plaintiffs alleged that statements about a proprietary health intelligence platform using advanced AI were materially misleading, as the platform reportedly lacked the claimed predictive capability.³¹ Zillow Group faced litigation after its algorithmic home-buying programme, Zillow Offers, failed to accurately predict housing market movements, leading the company to overpay for thousands of properties and shut down the business unit at a loss of approximately $528 million.³¹

These cases establish that when a company’s business model depends on the predictive accuracy of an algorithm, a wrong outcome is not merely a technical failure. It is a material risk requiring disclosure to investors.

Conclusion

The pattern across these sectors is consistent. Predictive analytics produce incorrect outcomes when they are used to manage complex human systems through reductive statistical averages. Whether the context is the denial of post-acute care, the undervaluation of homes in black neighbourhoods, or the automated extraction of welfare debts, the failures share common features. Loss of human context. Reliance on biased historical data. Absence of meaningful human oversight.

The litigation trends identified here, particularly the agency theory in employment law and the constitutional tort claims in public benefits, indicate that algorithmic impunity is ending. Organisations can no longer defend a wrong outcome by pointing to the algorithm. As regulatory frameworks such as the Colorado AI Act (effective 30 June 2026) come into force, the burden will increasingly fall on developers and deployers to demonstrate that their predictive models are accurate, equitable, and transparent.⁵

The technology is not the problem. The problem is deploying it without the accountability structures that the law requires for any system that determines individual rights and entitlements. Building those structures is the task ahead.

References

1. AI In Insurance Claim Review Raises Concerns Over Delays, Freedom for All Americans.

2. When Faulty AI Falls Into the Wrong Hands: The Risks of Erroneous AI, IJOC.

3. Lawsuit over AI usage by Medicare Advantage plans allowed to proceed, DLA Piper (2025).

4. AI Hiring Bias Lawsuits Are About to Surge, Reworked.

5. AI Bias in Hiring: Algorithmic Recruiting and Your Rights, Sanford Heisler Sharp (2025).

6. Two Lawsuits Expose AI Accountability Gaps in Hiring, Veris Insights.

7. When Machines Discriminate: The Rise of AI Bias Lawsuits, Quinn Emanuel.

8. AI Hiring Bias: Real Cases, Legal Consequences, and Prevention, Responsible AI Labs.

9. AI Hiring Tools Under Legal Scrutiny: Lessons for Employers, Hoyer Law Group.

10. Quantifying Losses for 443,000 Australians in a $1.2 Billion Robodebt Class Action, Vincents.

11. Robodebt not only broke the laws of the land, it also broke laws of mathematics, University of Wollongong (2023).

12. Why Robodebt failed, ANU Reporter.

13. Learning from the failures of Robodebt, Victoria Legal Aid.

14. Managing unintended consequences of algorithmic decision-making: The case of Robodebt, Sage Journals.

15. Michigan Unemployment Insurance False Fraud Determinations, BTAH.

16. Case Over the Michigan Unemployment Insurance Agency’s Faulty Automated System Finally Settled, Ford School (2024).

17. Automated Stategraft: Faulty Programming and Improper Collections in Michigan’s Unemployment Insurance Program, Wisconsin Law Review.

18. Michigan’s MiDAS Unemployment System: Algorithm Alchemy Created Lead, Not Gold, IEEE Spectrum.

19. The MIDAS Touch: Atuahene’s “Stategraft” and Unregulated AI, UNM Digital Repository.

20. EPIC v. DOJ (Criminal Justice Algorithms), EPIC.

21. DOJ Report on AI in Criminal Justice: Key Takeaways, Council on Criminal Justice.

22. Police departments sued over predictive policing programs, Police1.

23. The Dangers of Unregulated AI in Policing, Brennan Center for Justice.

24. The Legal Risks of Big Data Policing, American University Law.

25. Do Automated Valuation Models Reinforce Disparities in Home Values?, Urban Institute.

26. Racial Disparities in Automated Valuation Models: New Evidence Using Property Condition and Machine Learning, HUD.

27. Don’t trust AI for home valuation, estate agents warn, Mortgage Professional America.

28. Are AVMs Sabotaging Property Valuations?, Seaport Real Estate Services.

29. When Algorithms Judge Your Credit: Understanding AI Bias in Lending Decisions, Accessible Law, UNT Dallas.

30. When Algorithms Deny Loans: The Fraught Fight to Purge Bias from AI, IoT For All.

31. Consequences Clear for Firms that are AI-Washing, Labaton.

Courts are slow. That is neither a new observation nor a controversial one. Case backlogs across major jurisdictions have been mounting for decades and the cost of commercial litigation has risen in step with the complexity of the disputes feeding it. Artificial intelligence offers a set of responses to this problem. Some are administrative and uncontroversial. Others amount to a transfer of adjudicative authority from human judges to algorithmic systems. The distinction matters, because the efficiency gains of the first category do not justify the constitutional risks of the second.

This article examines the current state of AI deployment across civil and commercial courts globally, evaluates the legal, ethical, and economic consequences of that deployment and identifies the regulatory frameworks attempting to govern it. The central argument is that while assistive AI has already demonstrated measurable value in court administration, the move toward automated adjudication raises questions about due process, accountability and judicial legitimacy that remain unanswered.

Assistive AI and the Administrative Layer

The integration of AI into courtrooms operates on a spectrum. At the lower end sit administrative tools designed to relieve clerical bottlenecks. These are the least controversial applications and, in many jurisdictions, the most advanced.

Clerical Automation

China’s Supreme People’s Court has been the most aggressive adopter. Under the “Smart Courts” initiative, every Chinese court is expected to deploy AI tools to support judicial functions by 2030, with full embedding targeted across all tiers.^3,4 Optical character recognition, automatic speech recognition and natural language processing handle transcription, document classification, and procedural scheduling. Official figures claim a 30% reduction in average trial times, though independent verification of that number remains limited.³

Egypt has followed a similar path with AI-driven transcription tools as part of its judicial digitalisation programme.³ The immediate benefit is plain. In jurisdictions where manual transcription delays the progression of cases from trial to appeal, automated transcription removes an administrative chokepoint.

Document Review and Case Management

At a more sophisticated level, AI is handling the volume problem in commercial litigation. Singapore’s International Arbitration Centre has integrated AI into its cloud-based case management platform, the SIAC Gateway.⁵ One reported instance involved an AI tool reviewing over 750,000 documents in four weeks by identifying similar terms and clauses. The equivalent manual exercise would have required months of associate time.⁵

Brazil’s SIGMA system goes further. It assists judges in drafting decisions by analysing stored texts and procedural documents to identify relevant precedent, then suggests models and templates for reports and judgments.³ The stated objective is consistency. Like cases should receive like treatment, and in commercial law that is not merely a principle of fairness but a condition of market predictability.

Automated Adjudication and the Robot Judge

The jump from administrative assistance to decision-making is qualitative, not merely quantitative. A system that transcribes a hearing is performing a clerical function. A system that determines liability or issues an enforceable order is exercising judicial authority. Several jurisdictions have already crossed that line.

Small Claims and High-Volume Disputes

Estonia is the most frequently cited example. The Ministry of Justice has implemented a semi-automated procedure for small monetary claims. Computer-generated payment orders are issued automatically based on information supplied by the parties through the national e-File system, and these orders carry the legal status of judgments for enforcement purposes.^9,10 Human oversight is retained for jurisdictional determinations, but the core adjudicative output is algorithmic. The logic is to clear the backlog so that human judges can concentrate on cases that require human judgment.

British Columbia’s Civil Resolution Tribunal operates on a similar basis. Decision trees and expert systems resolve small claims up to C$5,000 and motor vehicle injury disputes up to C$50,000.¹¹ Many disputes conclude before a human adjudicator is ever involved. The model works for cases where the legal questions are binary and the stakes do not engage fundamental rights or liberty interests.

Predictive Analytics and Sentencing

In the United States, the application is different in character. Algorithms are used to predict case outcomes, assisting attorneys in advising clients on settlement versus trial.¹² More controversially, risk assessment tools such as COMPAS compute thousands of data points to generate scores that inform sentencing and parole decisions.² The data inputs range from criminal history to local crime rates to, in some formulations, proxies that correlate with race and socioeconomic status.

The American Arbitration Association announced in late 2025 a hybrid model in which AI generates draft outcomes for construction disputes, with human arbitrators reviewing and validating the result.¹³ The boundary between human and algorithmic adjudication is, in practice, already blurred.

Due Process and the Black Box

The constitutional objection to automated adjudication is not abstract. It turns on a concrete requirement. Judgments must be reasoned. A party who loses must be able to understand why.¹⁴ Deep learning models, and particularly neural networks, do not satisfy this requirement. Their internal logic is opaque, often to their own developers.^15,16

Transparency

If a court delegates decision-making to an opaque system, the losing party is deprived of the ability to identify error, challenge reasoning, or mount a meaningful appeal.¹⁷ In 2024, a Texas appeals court reportedly overturned a conviction on the basis that AI-generated evidence lacked the transparency required for adversarial testing. The defence argued, with success, that cross-examination of an algorithm is not possible in any meaningful sense.¹ The case, if accurately reported, illustrates a tension that will recur across jurisdictions. The adversarial system depends on the ability to test evidence, and proprietary algorithmic models resist testing by design.

Legal Stasis and Automation Bias

There is a second, less discussed risk. Law evolves. It is supposed to. Judges distinguish, overrule, and extend precedent in response to changing circumstances. AI systems trained on historical data are retrospective by definition.¹⁸ They optimise for consistency with what has come before, not for the creative interpretation that legal development sometimes requires.

The related problem is automation bias. A judge presented with an algorithmically generated risk score or draft judgment may find it difficult to deviate, particularly if the methodology behind the score is not understood.^15,18 The practical effect is a transfer of discretion from the bench to the developer, a transfer of power that occurs without democratic mandate or constitutional authority.^18,19

Accountability

When a judge errs, the remedy is appeal. When an algorithm errs, the question of responsibility becomes diffuse. The judge who relied on the output, the vendor who supplied the software and the data scientist who trained the model all occupy different positions in the chain of causation.¹⁵ A 2025 Dutch decision reportedly apportioned 80% of liability for an algorithmic error causing wrongful eviction to the software developer.¹ The precedent, if it holds, suggests that developers cannot disclaim responsibility for downstream judicial harm. But the case also exposes the absence of a settled framework for algorithmic liability in adjudicative contexts.

Bias, Fairness, and the Limits of Algorithmic Neutrality

The bias problem is well documented. AI systems trained on historical data will reproduce the biases embedded in that data.^19,20 The question is whether the legal system is prepared to confront this at the point of deployment rather than after the damage is done.

Encoded Discrimination

The most cited example remains COMPAS. ProPublica’s 2016 analysis found that Black defendants who did not go on to reoffend were classified as higher risk 45% more frequently than white defendants in equivalent circumstances.¹ The finding was contested by Northpointe (the system’s developer), but the statistical disparity has not been satisfactorily explained away.

More recent research has identified a subtler form of encoded bias. AI models have been found to rate African American English as more “toxic” than standard American English, even where the semantic content is identical.^1,21 This is not a training error in the conventional sense, it is a consequence of models internalising sociolinguistic hierarchies present in their training corpora. For judicial applications, where the language of a litigant or witness may influence an algorithmic assessment, this is not a theoretical risk.

The Human Element

Adjudication requires more than logical consistency. It requires judgment in the fuller sense, which includes the capacity to perceive remorse, to weigh context, to exercise mercy where the law permits it.¹⁵ These are not features that can be specified in a model. They are attributes of human cognition and conscience.

The concern is that automated systems treat litigants as data points rather than persons, and that human adjudicators, over time, become signatories to algorithmic outputs rather than authors of reasoned decisions.^14,18 The CEPEJ European Ethical Charter addresses this directly, insisting that AI must remain a tool in the service of justice and that final authority must rest with human judges.²²

Economics, Efficiency Gains, and Market Distortion

The Scale of the Opportunity

The economic case for AI in legal practice is strong on the numbers. A 2025 survey estimated that AI tools could save each US lawyer approximately 200 work hours per year, translating to roughly $20 billion in annual savings across the US legal market.^24,25 In high-volume litigation, specific tasks that previously consumed 16 hours of associate time have been reduced to minutes.²⁷

The Billable Hour Under Pressure

The billable hour model, which still accounts for over 80% of fee arrangements at large firms, is structurally incompatible with AI-driven efficiency.²⁷ If a task that previously took ten hours now takes one, the firm either charges less or captures the surplus through alternative pricing. The transition is already underway. Fixed-fee arrangements are expanding, and firms that invested early in AI infrastructure are pricing competitors out of commoditised work.²⁴

The distributional effects are uneven. Large firms with the capital to invest millions in AI infrastructure will absorb the transition. Mid-sized firms, which lack that capital but also lack the agility of AI-native boutiques, face compression from both ends.²⁷

Public Court Systems and the Cost Problem

For public courts, the economics are less favourable. California’s Court Case Management System carries a projected cost of $1.9 billion, a figure that many consider understated.²⁸ The US Federal Judiciary’s cybersecurity and IT modernisation plan for 2022–2027 is budgeted at approximately $440 million.²⁹ These are large sums for systems that will require continuous investment to avoid obsolescence within years of deployment.

Public Trust and Social Acceptance

The NCSC’s 2024 and 2025 polling data provides a useful baseline. Public trust in US state courts sits at 62-63%.^30,31 On AI specifically, the picture is more cautious. Fifty-one per cent of respondents believe AI will increase the risk of mistakes that human judges might not catch. Only 31% believe AI will improve court efficiency.³²

There is support for limited applications. Sixty-three per cent endorse using AI for FAQs and document translation.³¹ The resistance is concentrated around adjudicative functions.

The generational data is worth noting. Voters aged 18-29 are nine points more likely than older cohorts to describe state courts as “innovative.”³⁰ More unexpectedly, studies have found that Black participants express greater trust in AI-augmented judicial decisions than white participants.²¹ The hypothesis, which warrants further investigation, is that communities with historical experience of judicial bias may view algorithmic consistency as an improvement over human discretion, even acknowledging the risk of encoded bias.

Regulatory Frameworks

The EU AI Act

The EU AI Act classifies AI systems used in the administration of justice as “high-risk.”^33,34 This classification triggers obligations including quality management systems, technical documentation, automatic logging for at least six months, and Fundamental Rights Impact Assessments before deployment.^33,35,36

The OECD Principles and CEPEJ Charter

The OECD’s five principles for AI governance, adopted by the G20 and incorporated into both the EU AI Act and the NIST AI Risk Management Framework, establish the baseline international consensus.^37,38 The emphasis is on augmentation rather than replacement. AI should enhance human capability, particularly in sensitive domains.³⁹

The CEPEJ Charter, adopted in 2018, was the first European instrument to address AI in judicial systems specifically.^22,40 Its five principles (respect for fundamental rights, non-discrimination, quality and security, transparency, and user control) are broadly aligned with the OECD framework. The “under user control” principle is the most operationally significant. It requires that judges are not bound by AI outputs and retain authority to review and override any algorithmic recommendation.^41,42

Hybrid Models and Strategic Positioning

The future of judicial AI is not replacement. No jurisdiction of any consequence is proposing to remove human judges from the determination of contested matters. The operative model is hybrid. AI handles volume, identifies patterns, and drafts preliminary outputs. Humans exercise judgment, apply discretion, and bear responsibility.⁷

Strategic Hubs

Certain jurisdictions are positioning themselves as centres of AI-enabled commercial justice. The DIFC has launched a five-year strategy (2026–2030) to become the most advanced international commercial court system, including a specialised Digital Economy Court for disputes involving big data, blockchain, and fintech.^7,43,44 Singapore is testing generative AI tools in the Small Claims Tribunal to help self-represented litigants understand opposing positions and evidence.⁴⁵

Professional Responsibility

As AI tools become more capable, the professional duties of lawyers and judges shift accordingly. Mata v. Avianca, Inc. remains the cautionary marker, with attorneys sanctioned for submitting fictitious AI-generated case citations to a federal court.^5,15 The lesson is not that AI should be avoided, but that its outputs require verification with the same rigour applied to any other source. The professional standard is competence, and competence now includes understanding the tools.

Conclusion

The efficiency gains from assistive AI in court administration are real and, in most cases, welcome. Automated transcription, document review, and case management tools reduce cost and delay without raising constitutional concerns. The more difficult questions arise where AI assumes adjudicative functions, because efficiency is not the only value the legal system serves.

Due process requires transparency. Accountability requires identifiable decision-makers. Fairness requires that the biases embedded in historical data are confronted rather than automated. No regulatory framework currently in force fully addresses these requirements in the context of judicial AI.

The task for the next decade is to build that framework. The CEPEJ Charter, the EU AI Act, and the OECD Principles provide a starting point. But principles are not enforcement mechanisms, and enforcement mechanisms are not yet calibrated to a world in which software determines enforceable legal rights. The technology is ahead of the governance. That gap is where the risk sits.

References

1. AI in Judicial Decisions: 5 Stunning Breakthroughs Redefining Courts in 2025, Medium (2025).

2. Robot Judgements In The Courtroom: What Do You Think?, The Open University.

3. AI in Global Majority Judicial Systems, Stimson Center (2026).

4. China | AI deeply embedded in criminal justice system, Oxford Institute of Technology.

5. AI in Arbitration – A Perspective from Singapore, Clyde & Co (2025).

6. Smart AI Tools will Transform How SG Firms Handle Legal Research, IMDA.

7. DIFC Courts launches new five-year Growth Strategy.

8. Technology, AI and the Future of Litigation in Dubai’s Onshore Courts, BSA Law.

9. Estonia | AI for crime prevention, Oxford Institute of Technology.

10. AI Robot Judge for small claims in Estonia, IPS-X.

11. From Courtroom to Codebase: Are You Ready for AI in Adjudication?, Today’s Managing Partner.

12. AI in the Courtroom, UCR Extension.

13. The new “Code” of conduct in resolving disputes: UAE’s AI vision, Al Tamimi & Company.

14. Commercial Dispute Resolution and AI (Chapter 23), The Cambridge Handbook of Private Law and Artificial Intelligence.

15. AI in Judicial Decision Making: Transparency & Ethics, Justice Speakers Institute.

16. The Legal And Ethical Implications Of Artificial Intelligence In The Judiciary, IJLLR.

17. Artificial Intelligence and Procedural Due Process, Penn Carey Law.

18. AI in the Courtroom: The Boundaries of RoboLawyers, Fordham IPLJ.

19. Ethical Challenges of Using Artificial Intelligence in Judiciary, arXiv (2025).

20. Avoiding Algorithmic Bias: Top 5 AI Liability Issues in Courts, Super Lawyers.

21. Public Perceptions of Judges’ Use of AI Tools in Courtroom Decision-Making, PMC.

22. CEPEJ European Ethical Charter on the use of AI in judicial systems, Council of Europe (2018).

23. JudgeGPT: The Benefits and Challenges of an AI Judiciary, NAPCO (2025).

24. Future of Professionals report analysis: Why AI will flip law firm economics, Thomson Reuters.

25. AI Usage Could Save U.S. Legal Industry $20 Billion Annually, 2Civility.

26. Financial Impacts of AI, American Bar Association (2023).

27. The Impact of Artificial Intelligence on Law Firms’ Business Models, Harvard CLP.

28. ‘Finished’ Court IT Project to Cost State 100s of Millions for Years, Courthouse News.

29. Judiciary Information Technology Fund, United States Courts.

30. State of the State Courts: 2025 public opinion poll findings, NCSC.

31. New Poll Shows Public Trust in State Courts on the Rise, Judicature.

32. NCSC Survey Finds Public Trust in State Courts Remains Strong Amid Emerging Concerns About AI, 2Civility.

33. Zooming in on AI – EU AI Act: obligations for high-risk AI systems, A&O Shearman.

34. High-level summary of the AI Act, EU Artificial Intelligence Act.

35. What Are High-Risk AI Systems Within the Meaning of the EU’s AI Act?, WilmerHale.

36. Article 16: Obligations of Providers of High-Risk AI Systems, EU AI Act.

37. OECD AI Principles, AI Ethics Lab.

38. Global AI Governance: Five Key Frameworks Explained, Bradley (2025).

39. AI principles, OECD.

40. European Ethical Charter on AI in judicial systems: implications, UNIO.

41. Artificial Intelligence and Judicial Activities: The Position of CEPEJ, IAJ-UIM.

42. European Ethical Charter on the Use of AI in Judicial Systems, full text.

43. Harnessing innovation for future dispute resolution, DIFC Courts.

44. DIFC Courts unveils five-year strategy prioritising AI, Law Middle East.

45. Media Release: New Generative AI-powered Case Summarisation Tool, Singapore Courts.

46. Out in the Dark with Artificial Intelligence, SAL Practitioner.

47. AI in justice administration and access to justice, OECD.

48. SC 42 – JTC 1.

Artificial intelligence is already inside the machinery of policing, but not in the way public debate often suggests. The most important uses are usually administrative and investigative rather than theatrical. Systems are used to search images, rank risk, identify patterns in incident data, draft reports, redact documents and sort large quantities of digital material. None of that is futuristic. It is current practice. The legal question is not whether software now assists law enforcement. The real question is what happens to privacy, equality, disclosure and due process when suspicion is generated, refined and acted upon through systems that are marketed as efficient but are not always transparent, contestable or understood by the institutions using them. ¹It is a mistake to speak about “AI in law enforcement” as though it were one technology and one legal problem. A system that transcribes body-camera audio and generates a draft narrative does not raise the same concerns as live facial recognition in a public street. A retrospective image-matching tool used after an incident is not the same as an automated risk model that affects bail, sentencing or watchlisting. Nor is an internal triage system the same as forensic software tendered in criminal proceedings. The law tends to become confused when these categories are collapsed into a single argument about innovation or danger. The more disciplined approach is to separate administrative uses, investigative uses, surveillance uses and evidential uses and then ask what safeguards are required at each level of legal consequence. ²

That distinction matters because policing is not an ordinary market activity. It involves the organised use of public power. When the state watches, categorises, searches, arrests or prosecutes, it acts against a legal background that is meant to restrain convenience. That is why the central issues in this field are still familiar legal ones. What is the statutory or common law basis for using the tool? What is the threshold for necessity and proportionality? What information can the affected person obtain? Can the system be independently tested? Can an officer, prosecutor or judge explain the role the output played in the decision? Once those questions are asked properly, much of the marketing language around artificial intelligence falls away. The real issue is not whether the state is “modernising”, but whether the conditions for lawful coercive power remain intact. ³

I. The UK Position: Principle, Litigation and Legislative Catch-Up

The United Kingdom has so far preferred a sectoral and principles-led model rather than a single, horizontal AI statute. That approach has a certain practical logic. It avoids drafting one abstract code for technologies that perform very different functions in very different institutional settings. But it also means that some of the real regulation happens after deployment, through challenge, guidance and litigation, rather than before deployment through bright-line statutory rules. In fields as sensitive as policing, that is a material choice. It leaves police forces and regulators to assemble legality from a mixture of human rights law, data protection law, equality law, common law policing powers and internal operational guidance. Sometimes that is enough. Sometimes it is not. ⁴

The leading authority is still Bridges. The significance of the case lies in its refusal to allow broad operational confidence to stand in for legal discipline. The Court of Appeal held that South Wales Police’s use of automated facial recognition on the facts was unlawful. The deployment was not “in accordance with the law” for the purposes of article 8(2) of the Convention because too much was left to police discretion. The force’s Data Protection Impact Assessment was also deficient and its treatment of the public sector equality duty was inadequate. The judgment matters not because it outlawed facial recognition in principle, but because it explained that intrusive biometric surveillance requires a framework that is clear enough to constrain who can be watched, where the technology can be used and how discriminatory effects are to be examined before confidence is asserted. ⁵

That point is still current. In December 2025 the Home Office accepted, in terms, that the present legal position is “complicated, inflexible and difficult to understand” and opened a consultation on a new framework for law enforcement use of facial recognition and related technologies. The consultation does not proceed from the premise that nothing is lawful now. Rather, it proceeds from the more important premise that legality assembled from overlapping powers and guidance is not an ideal basis for public trust or confident operational expansion. The Government’s own language is revealing: the law needs to keep pace with technological developments and provide clear, consistent rules that the public can understand more easily and that law enforcement can rely upon. That is a polite way of recognising that the existing architecture is serviceable only up to a point. ⁶

The Gangs Matrix provides a different lesson. It was not generative AI and not all of its functions would satisfy every modern definition of artificial intelligence. But that does not alter the legal significance of the system. It was still a state-run classificatory tool used to assign levels of perceived risk to individuals, with practical consequences beyond the police database in which the scores were stored. The ICO’s intervention established that serious data protection failings existed and the Metropolitan Police has now discontinued the Gangs Violence Matrix entirely. What makes the episode important is not the branding of the tool but the logic behind it: once risk-scoring systems are embedded in policing and shared across agencies, they can shape housing decisions, employment prospects, school interventions and surveillance intensity without any meaningful hearing ever taking place. ⁷

The real vice in systems of that kind is not simply that they may be wrong. All policing information can be wrong. The deeper problem is that they distribute administrative stigma in a way that is difficult to see and difficult to challenge. A person does not need to know the model architecture to be harmed by being incorrectly marked as high risk. What matters is that the classification may affect how public and quasi-public bodies behave towards him before any court has tested the basis of that classification. In public law terms, this is where data quality, retention, necessity and fairness cease to be technical questions and become constitutional ones. The fact that software has assisted the result does not reduce the seriousness of the decision. It often increases it by obscuring how the decision took shape. ⁸

II. The United States: Due Process, Disclosure and Wrongful Arrest

The United States approaches many of the same problems through a different legal vocabulary. There the pressure points are usually due process, confrontation, probable cause, evidential admissibility and disclosure of proprietary methods. The underlying issue, however, is much the same as in the UK. How much opacity can the criminal process tolerate before the process ceases to be fair? The question arises in several settings: sentencing tools, probabilistic genotyping software, facial recognition, AI-enhanced video and other systems that convert contested inputs into outputs carrying an aura of technical authority. The hard cases are those in which a defendant is materially affected by a system whose functioning the defence cannot properly test. ⁹

State v Loomis is the obvious starting point. Correctional Offender Management Profiling for Alternative Sanctions, ‘COMPAS’ is a proprietary risk assessment tool used in US criminal proceedings to generate scores predicting a defendant’s likelihood of reoffending, used to inform bail, sentencing and parole decisions. The Wisconsin Supreme Court did not reject the use of COMPAS, but nor did it embrace algorithmic authority in the broad sense sometimes implied in commentary. It held that a sentencing court could consider the assessment subject to strong limitations and warnings. That is important because the case is often invoked as though it settled the legitimacy of proprietary risk tools in criminal justice. It did not. It tolerated a constrained use and did so against a background of explicit concern about the proprietary nature of the system and its use of gender. The unease remained in the judgment itself. A process in which liberty may be affected by an assessment generated by a tool the defence cannot meaningfully inspect is not made easy merely by saying that the output is only one factor among others. ¹⁰

The same concern is sharper when software moves from sentencing support to forensic attribution. In New Jersey, State v Pickett and the later discussion of Pickett in State v Rochat show why source code scrutiny matters in criminal proceedings. The point is not that every defendant should automatically receive the source code of every proprietary program used somewhere in the investigative chain. The point is that claims of trade secrecy cannot always defeat the demands of a fair criminal process. The New Jersey cases treat the issue as one of reliability, adversarial testing and the court’s own responsibility to scrutinise novel scientific evidence carefully. That is a healthier way to frame the problem than treating forensic software as presumptively trustworthy because it has been sold to prosecutors and validated by its maker. ¹¹

That line of authority matters beyond DNA evidence. It expresses a simple procedural principle: if software does work that bears directly on guilt, causation, identification or sentence, the legal system must be able to test its reliability in a way that is more than ceremonial. Courts cannot sensibly perform a gatekeeping function while being told, in effect, that the method is too commercially sensitive to examine. Nor can defendants mount a serious challenge if they are forced to accept vendor assertions as a substitute for adversarial scrutiny. In that sense, the “black box” problem is not mainly philosophical, it is procedural. It concerns the ability of a court to know enough about the method to judge whether it is fit to be used in a process that may end in conviction or imprisonment. ¹²

Facial recognition makes the human cost of this procedural weakness easier to see. The wrongful arrest of Robert Williams in Detroit is now the emblematic American example and it remains powerful because it stripped away any abstract talk of efficiency. Williams was arrested for a theft he did not commit after the police relied on an incorrect facial-recognition result generated from poor-quality footage. The resulting litigation led to a 2024 settlement imposing what the ACLU described as the strongest police-department guardrails in the country. The settlement is important not because it abolished facial recognition, but because it imposed a principle the law should probably have insisted upon from the start: a facial-recognition lead cannot by itself justify an arrest and even an ensuing lineup is not independent if it simply launders the original algorithmic suggestion. ¹³

That settlement is part of a broader American pattern. The NCSL survey records that states including Alabama, Maryland and Washington have introduced or enacted measures requiring accountability reports, judicial authorisation in some contexts, or corroborating evidence before law enforcement use can translate into coercive consequences. The detail varies from state to state, but the instinct is the same. Legislatures have begun to recognise that algorithmic resemblance is not probable cause and that the appearance of computational precision can distort ordinary investigative scepticism. Once an officer feels he is corroborating a machine rather than testing a lead, error can harden into action very quickly. The point of guardrails is to slow that process down and force human investigation back into the chain. ¹⁴

There is a lesson here for English lawyers as well. The language of “human in the loop” is too thin if it simply means a person presses the final button. Human involvement is only a safeguard if the human actor is trained, sceptical and empowered to reject the output rather than ratify it. Otherwise, human review becomes a ceremonial stage in an automated process. That is why the legal significance of AI cannot be measured simply by asking whether the final decision is formally made by a person. The better question is whether the institutional design encourages independent judgement or whether it encourages a passive acceptance of machine-generated conclusions, particularly where those conclusions arrive dressed in the language of risk scores, confidence metrics or biometric similarity. ¹⁵

That same issue appears in a more mundane but increasingly important part of policing: report generation and administrative drafting. Axon’s Draft One, for example, uses body-worn camera audio to generate a draft report narrative. Bedfordshire Police has publicised the use of artificial intelligence to auto-redact documents before disclosure. These uses may look relatively benign when compared with public-space facial recognition or forensic software and in one sense they are. But they should not be treated as legally irrelevant. Drafting tools shape sequence, tone, omission and emphasis. Redaction tools shape what prosecutors, defence lawyers and courts are able to see. When such systems are adopted for efficiency reasons, there is a risk that the administrative record comes to reflect the logic of the software rather than the judgement of the officer or reviewer supposedly responsible for it. ¹⁶

None of this is an argument against administrative assistance. Police forces plainly waste vast amounts of time on routine bureaucracy and some of that work can sensibly be reduced. The point is narrower and more legal. The closer a system moves to the creation of evidence, the framing of a witness account, the presentation of disclosure material or the explanation of an arrest, the more unrealistic it becomes to describe the product as merely clerical. Once a generated text is reviewed, amended and signed, it enters the legal bloodstream. At that point the law has to care about how it was produced, what source material it used, what it omitted and whether the human signatory is in any meaningful sense the author of the resulting narrative. Administrative convenience is not a defence to later evidential confusion. ¹⁷

III. Administrative Failure, Comparative Regulation and the Market for Enforcement Technology

The broader administrative warning comes from outside criminal enforcement itself. Australia’s Robodebt scheme has become the clearest illustration of what happens when an automated process is allowed to outrun legal control inside the state. The Robodebt scheme was an automated debt recovery system operated by the Australian government between 2016 and 2019, which used income averaging to generate welfare overpayment notices, a method later found to be unlawful, resulting in a $1.8 billion settlement and a Royal Commission. The scandal was not about robots making sovereign decisions. It was about officials becoming too willing to trust an automated logic because it appeared scalable, consistent and technically grounded. The Royal Commission’s report is especially valuable because it is not written in the language of speculative ethics. It is a practical account of public administration going wrong. Its recommendations call for a consistent legal framework for automation in government services, for review rights, for public information about automated systems, and for publication of business rules and algorithms sufficient to allow independent scrutiny. That is not anti-technology. It is a restatement of ordinary administrative law after a conspicuous failure of institutional judgement. ¹⁸

Robodebt belongs in any serious analysis of AI and policing because it demonstrates a pattern that is not confined to welfare administration. The danger is not only a technical defect, it is the institutional tendency to treat process outputs as presumptively sound because they emerge from a system rather than a person. Review pathways then weaken. Discretion is compressed. Affected individuals are left to challenge a conclusion after it has already been operationalised against them. In policing, the consequences may be different, but the pattern is recognisable. Risk tools, watchlists, facial-recognition alerts and AI-assisted summaries can all become dangerously authoritative when they are absorbed into workflow without a corresponding increase in scepticism and accountability. ¹⁹

The European Union has responded in a more prescriptive way than the UK. The AI Act proceeds by risk classification and treats many law enforcement systems as high risk. More importantly, it prohibits certain practices outright. The political significance of that approach is easy to miss. The Act does not merely ask public authorities to use judgment more carefully. It draws legislative lines in advance. Among the practices prohibited are untargeted scraping of facial images from the internet or CCTV to create facial-recognition databases and certain forms of predictive policing based solely on profiling or assessments of personality and personal characteristics. The use of real time remote biometric identification in publicly accessible spaces for law enforcement purposes is prohibited in principle, save for narrow and tightly conditioned exceptions. ²⁰

Whatever one thinks of the detail, the Act makes an important constitutional choice. It assumes that some uses of artificial intelligence in law enforcement are too rights-sensitive to be governed mainly through soft guidance and ex post litigation. The UK has made a different choice, or at least has done so thus far. It has preferred common law powers, data protection law, equality law, human rights law and operational guidance, with fresh legislation now under consultation. The EU model risks rigidity and definitional complexity. The UK model risks ambiguity, uneven practice and a dependence on litigation to clarify boundaries. Those are different legal temperaments, not merely different drafting styles. For policing, the question is which model better protects the public against routine normalisation of intrusive tools before democratic scrutiny catches up. ²¹

The AI Act also matters because it speaks directly to the language of explanation and contestability. Commentary often says loosely that there is now a broad “right to explanation” in relation to AI. That overstates the position. The better view is that data protection law and the AI Act create a set of more specific rights and obligations relating to transparency, review and information about the logic involved, particularly where significant effects or high-risk uses are concerned. The legal point is not to promise a perfect decoding of every technical model. It is to ensure that a person subject to a materially significant automated or AI-assisted decision is not left with nothing more than the assurance that the system has been designed responsibly. In legal process, trust has to be accompanied by reasons, records and routes of challenge. ²²

Clearview AI is a US-based company that built a facial recognition database by scraping billions of publicly available images from the internet and social media without the knowledge or consent of the individuals depicted and sold access to law enforcement and security agencies worldwide. It shows the problem from another angle. European regulators have repeatedly taken the view that the company’s business model is incompatible with privacy law. The Italian authorities imposed a €20 million fine, ordered erasure and banned further collection and processing concerning persons in Italy. What makes Clearview important is not only the scale of the database but the premise of the enterprise: publicly available images were converted into a biometric search infrastructure for law enforcement and security customers without the consent, and usually without the knowledge, of the people whose faces were scraped. From a rights perspective, that is a profound repurposing of information. From a regulatory perspective, it is a test of whether domestic privacy law can control an international company offering tools to police and security agencies from outside the jurisdiction. ²³

IV. Evidence, Oversight and the Limits of Administrative Convenience

The United Kingdom’s encounter with Clearview has illustrated the jurisdictional difficulty rather than removing it. The 2025 Upper Tribunal decision revisited the reach of UK data protection regulation where the company’s clients operated in the fields of national security or criminal law enforcement. For present purposes, the significance of the case is twofold. First, it shows that domestic enforcement against cross-border biometric providers will often turn on difficult questions of territorial scope and regulatory competence rather than on the ethics of scraping alone. Secondly, it confirms that the growth of AI-enabled law enforcement infrastructure increasingly depends on private providers whose accountability cannot be assumed simply because the purchasing agency is public. The constitutional problem is therefore not only how the police use the technology, but also what sort of technology market the law is willing to permit around them. ²⁴

That leads to liability and procurement. Public discussion tends to focus on whether an individual officer made the final decision. In practice, many of the most important choices are made much earlier: when a force buys a system, accepts a validation claim, agrees a data architecture, signs a licence, limits audit access or relies on a vendor’s assertion about bias testing. If those choices are made badly, later “human review” may not rescue the process. The public body will still be the constitutional actor, but that does not make the vendor legally or practically irrelevant. Once software becomes part of the operational chain, questions about warranties, audit rights, indemnities, explainability, retention and downstream use are not mere contract management. They are part of the legality of the system as deployed. ²⁵

English public lawyers may not need a new grand theory to see the point. The relevant principles are already familiar. A public body exercising coercive functions must know what it is doing, must be able to justify why it is doing it, and must retain responsibility for the result. If it purchases a system it cannot explain, cannot properly test, and cannot effectively audit, it is not made safer by the fact that a private supplier claims confidence in the product. If anything, the risk is greater. The harder it is to examine the method, the easier it becomes for responsibility to dissolve into process. That is the real constitutional danger of AI in law enforcement. It is not machine consciousness or autonomous intent. It is institutional diffidence in the face of vendor claims and internally generated technical mystique. ²⁶

The evidential dimension deserves separate emphasis because criminal process does not merely use technology; it legitimises it. Once a court admits an AI-assisted output, or allows an official narrative shaped by automated drafting to stand without real scrutiny, the system acquires institutional authority far beyond its immediate function. That is why ordinary evidential principles matter so much here. Reliability, disclosure, expert challenge and the judge’s gatekeeping role are not technical side issues. They are the mechanisms by which the legal system protects itself from mistaking industrialised convenience for proof. The point is especially sharp where the software does not just organise material but adds interpretive weight to it, whether by ranking probabilities, highlighting “matches”, generating summaries or suppressing what it treats as irrelevant. In such cases the method becomes part of the evidence even if the operator insists the machine only assisted. ²⁷

That concern is no longer confined to specialist courts or complex forensic disputes. The senior judiciary in England and Wales has now repeatedly warned that AI-generated material cannot be relied upon uncritically and that judicial office holders remain responsible for everything issued in their name. Although that guidance is directed to judges, the underlying proposition travels more widely. Responsibility does not evaporate because a system appears useful. If anything, systems that appear useful require greater caution, because they invite routine dependence. In policing, that means supervisors, investigators and disclosure officers need enough understanding of the tools they use to recognise where the output may be incomplete, distorted or overconfident. Without that literacy, “human oversight” becomes a phrase of reassurance rather than a working safeguard. ²⁸

Conclusion

Public trust is often invoked by police authorities as a policy objective, but it should not be treated as something achieved by communication strategy alone. The Home Office’s own work on public attitudes to facial recognition shows a more nuanced picture: many members of the public see clear benefits in locating suspects, finding missing people and improving public safety, but support is qualified by concern about misuse, false matches and privacy. That is exactly what one would expect in a constitutional democracy. Citizens do not need to reject the utility of a tool in order to insist upon strict limits governing its use. The right lesson for public authorities is therefore not that support exists, but that support remains contingent on visible safeguards, intelligible rules and confidence that the technology will not quietly migrate from exceptional use into routine surveillance. ²⁹

The temptation in this field is to ask whether the law is “keeping up” with technology. That is a useful question only if one remembers that the law’s function is not merely to keep pace, but to decide where pace is appropriate and where it is not. Some policing uses of AI are straightforward productivity tools. Others alter the conditions on which people can move anonymously in public space, become subjects of suspicion, or find themselves on the receiving end of an arrest or prosecution. The legal system does not owe each of those uses the same level of enthusiasm. It owes them differentiated scrutiny. A sensible framework would therefore distinguish clearly between low-risk administrative assistance, higher-risk investigative uses, public-space biometric surveillance and evidential software, with escalating standards of authorisation, auditability, human review and disclosure as the legal stakes rise. ³⁰

For the United Kingdom, that means the next phase should be less rhetorical and more institutional. If facial recognition and related technologies are to expand, then Parliament or the Government must say with precision what powers are being exercised, which technologies are covered, what authorisations are required, what data sources are legitimate, what testing is mandatory, what records must be kept, what disclosure obligations arise when the output is relied upon, and what remedies are available when the system is used unlawfully or inaccurately. A principles-led model can only take matters so far if the principles are not translated into durable operating rules. Bridges showed that the courts can supply some of the discipline. They cannot sensibly be expected to supply all of it. ³¹

The most defensible final position is neither prohibitionist nor complacent. Law enforcement should use technology where it genuinely improves legitimate policing objectives and where the legal framework is robust enough to support that use. But the burden belongs to the state. It must show more than functionality. It must show lawful basis, necessity, proportionality, testability, meaningful oversight and practical routes of challenge. Systems that cannot meet those standards should not be used merely because they are available or because procurement cycles and vendor presentations make adoption seem inevitable. In a legal order worth preserving, AI may assist the state, but it cannot dilute the standards by which the state is judged when it watches, classifies, accuses or restrains. ³²

Selected Authorities and Materials

ACLU of Michigan, Civil Rights Advocates Achieve the Nation’s Strongest Police Department Policy on Facial Recognition Technology (28 June 2024).

ACLU of Michigan, Facial Recognition case page.

Ayres, Ian and Jack Balkin, ‘The Law of AI Is the Law of Risky Agents Without Intentions’ (2024) The University of Chicago Law Review Online.

Axon, A closer look at Draft One (2025).

Axon, Draft One product page.

Bedfordshire Police, Cutting edge tech saving Bedfordshire Police officers’ time (29 December 2023).

Bellovin, Steven M. et al., ‘Seeking the Source: Criminal Defendants’ Constitutional Right to Source Code’ (2021) 17 Ohio State Technology Law Journal 1.

European Data Protection Board, Facial recognition: Italian SA fines Clearview AI EUR 20 million (10 March 2022).

European Data Protection Board, The French SA fines Clearview AI EUR 20 million (20 October 2022).

European Parliament, Artificial Intelligence Act: MEPs adopt landmark law (13 March 2024).

Home Office, Legal framework for using facial recognition in law enforcement (consultation, 4 December 2025).

Home Office, Police use of facial recognition: factsheet (4 December 2025).

Home Office, Public attitudes to police use of facial recognition technology (4 December 2025).

ICO, Metropolitan Police gangs matrix.

ICO, Rights related to automated decision making including profiling.

ICO, What is automated individual decision-making and profiling?.

Judiciary of England and Wales, Artificial Intelligence (AI) Guidance for Judicial Office Holders (31 October 2025).

Metropolitan Police, How the gangs violence matrix works.

National Conference of State Legislatures, Artificial Intelligence in Law Enforcement: The Federal and State Landscape (2024).

Regulation (EU) 2024/1689 (Artificial Intelligence Act).

Royal Commission into the Robodebt Scheme, Report (7 July 2023).

The Information Commissioner’s Office v Clearview AI Inc (Privacy International intervening): [2025] UKUT 319 (AAC).

R (Bridges) v Chief Constable of South Wales Police [2020] EWCA Civ 1058.

State v Loomis, 2016 WI 68.

State v Pickett, 466 N.J. Super. 270 (App. Div. 2021).

State v Rochat, A-0103-17 (App. Div., 28 January 2022).

References

¹ Home Office, Police use of facial recognition: factsheet (4 December 2025); Axon, Draft One.

² Information Commissioner’s Office, What is automated individual decision-making and profiling?.

³ See R (Bridges) v Chief Constable of South Wales Police [2020] EWCA Civ 1058, https://www.judiciary.uk/wp-content/uploads/2020/08/R-Bridges-v-CC-South-Wales-ors-Judgment.pdf; UK Information Commissioner’s Office, Rights related to automated decision making including profiling, https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/individual-rights/individual-rights/rights-related-to-automated-decision-making-including-profiling/.

⁴ UK Government, AI regulation: a pro-innovation approach (White Paper, March 2023), https://www.gov.uk/government/publications/ai-regulation-a-pro-innovation-approach; Home Office, Legal framework for using facial recognition in law enforcement (consultation, published 4 December 2025), https://www.gov.uk/government/consultations/legal-framework-for-using-facial-recognition-in-law enforcement.

⁵ R (Bridges) v Chief Constable of South Wales Police [2020] EWCA Civ 1058; see also Judiciary UK, Press Summary: R (Bridges) v Chief Constable of South Wales Police (11 August 2020), https://www.judiciary.uk/wp-content/uploads/2020/08/R-Bridges-v-CC-South-Wales-ors-Press-Summary.pdf.

⁶ Home Office, Legal framework for using facial recognition in law enforcement (consultation, published 4 December 2025), especially the Summary and Consultation Description; Home Office, Police use of facial recognition: factsheet (4 December 2025).

⁷ Information Commissioner’s Office, Metropolitan Police gangs matrix, https://ico.org.uk/for-the-public/ico-40/metropolitan-police-gangs-matrix/; Metropolitan Police, How the gangs violence matrix works, stating that use of the matrix was discontinued with effect from 13 February 2024, https://www.met.police.uk/police-forces/metropolitan-police/areas/about-us/about-the-met/gangs-violence-matrix/.

⁸ Information Commissioner’s Office, Metropolitan Police gangs matrix; UK Information Commissioner’s Office, Rights related to automated decision making including profiling; see also Home Office, Legal framework for using facial recognition in law enforcement (consultation, 2025), which expressly raises issues of safeguards, proportionality and rights interference.

⁹ National Conference of State Legislatures, Artificial Intelligence in Law Enforcement: The Federal and State Landscape (July 2024), https://documents.ncsl.org/wwwncsl/Criminal-Justice/Law enforcement-Fed-Landscape-v02.pdf; Wisconsin Supreme Court, State v Loomis, 2016 WI 68, https://www.wicourts.gov/sc/opinion/DisplayDocument.pdf?content=pdf&seqNo=171690.

¹⁰ State v Loomis, 2016 WI 68, especially the certified issue concerning whether use of a COMPAS assessment violated due process because of the tool’s proprietary nature or its use of gender; see also Wisconsin Court of Appeals certification order, 17 September 2015, https://www.wicourts.gov/ca/cert/DisplayDocument.pdf?content=pdf&seqNo=149036.

¹¹ State v Pickett, 466 N.J. Super. 270 (App. Div. 2021), available at https://www.njcourts.gov/system/files/court-opinions/2021/a4207-19.pdf; State v Rochat, A-0103-17 (App. Div., 28 January 2022), discussing Pickett and describing independent source-code review as justified by the cautionary experience with FST, https://www.njcourts.gov/system/files/court-opinions/2022/a0103-17.pdf.

¹² See State v Pickett, 466 N.J. Super. 270 (App. Div. 2021); Rebecca Wexler, It’s time to end the trade secret evidentiary privilege among forensic algorithm vendors, Brookings (13 July 2021), https://www.brookings.edu/articles/its-time-to-end-the-trade-secret-evidentiary-privilege-among-forensic-algorithm-vendors/; Steven M. Bellovin et al., ‘Seeking the Source: Criminal Defendants’ Constitutional Right to Source Code’ (2021) 17 Ohio State Technology Law Journal 1.

¹³ ACLU of Michigan, Facial Recognition case page, https://www.aclumich.org/cases/facial-recognition/; ACLU of Michigan, Civil Rights Advocates Achieve the Nation’s Strongest Police Department Policy on Facial Recognition Technology (28 June 2024), https://www.aclumich.org/press-releases/civil-rights-advocates-achieve-nations-strongest-police-department-policy-facial/.

¹⁴ National Conference of State Legislatures, Artificial Intelligence in Law Enforcement: The Federal and State Landscape (2024), discussing state measures in Alabama, Maryland and Washington; ACLU of Michigan, Civil Rights Advocates Achieve the Nation’s Strongest Police Department Policy on Facial Recognition Technology (28 June 2024).

¹⁵ Judiciary of England and Wales, Artificial Intelligence (AI) Guidance for Judicial Office Holders (31 October 2025), https://www.judiciary.uk/wp-content/uploads/2025/10/Artificial-Intelligence-AI-Guidance-for-Judicial-Office-Holders-2.pdf; UK Information Commissioner’s Office, Rights related to automated decision making including profiling.

¹⁶ Axon, Draft One product page; Axon, A closer look at Draft One (2025), https://www.axon.com/resources/closer-look-draft-one; Bedfordshire Police, Cutting edge tech saving Bedfordshire Police officers’ time (29 December 2023).

¹⁷ Axon, Draft One product page; Judiciary of England and Wales, Artificial Intelligence (AI) Guidance for Judicial Office Holders (31 October 2025), which stresses that judicial office holders remain personally responsible for material produced in their name.

¹⁸ Royal Commission into the Robodebt Scheme, Report (7 July 2023), especially Chapter 17 on automated decision-making and Recommendation 17.1, https://robodebt.royalcommission.gov.au/publications/report.

¹⁹ Royal Commission into the Robodebt Scheme, Report (2023); UK Information Commissioner’s Office, Rights related to automated decision making including profiling; Home Office, Legal framework for using facial recognition in law enforcement (consultation, 2025).

²⁰ Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 (Artificial Intelligence Act), OJ L 2024/1689, especially Articles 5 and 6, https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689; European Parliament, Artificial Intelligence Act: MEPs adopt landmark law (13 March 2024), https://www.europarl.europa.eu/news/en/press-room/20240308IPR19015/artificial-intelligence-act-meps-adopt-landmark-law.

²¹ European Parliament, Artificial Intelligence Act: MEPs adopt landmark law (13 March 2024); Home Office, Legal framework for using facial recognition in law enforcement (consultation, 4 December 2025).

²² UK Information Commissioner’s Office, Rights related to automated decision making including profiling and What else do we need to consider if Article 22 applies?, https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/individual-rights/automated-decision-making-and-profiling/what-else-do-we-need-to-consider-if-article-22-applies/; Regulation (EU) 2024/1689, including Article 86.

²³ European Data Protection Board, Facial recognition: Italian SA fines Clearview AI EUR 20 million (10 March 2022), https://www.edpb.europa.eu/news/national-news/2022/facial-recognition-italian-sa-fines-clearview-ai-eur-20-million_en; see also European Data Protection Board, The French SA fines Clearview AI EUR 20 million (20 October 2022), https://www.edpb.europa.eu/news/national-news/2022/french-sa-fines-clearview-ai-eur-20-million_en.

²⁴ Upper Tribunal (Administrative Appeals Chamber), The Information Commissioner’s Office v Clearview AI Inc (Privacy International intervening): [2025] UKUT 319 (AAC), published 14 October 2025, https://www.gov.uk/administrative-appeals-tribunal-decisions/the-information-commissioners-office-v-clearview-ai-inc-privacy-international-intervening-2025-ukut-319-aac.

²⁵ Home Office, Legal framework for using facial recognition in law enforcement (consultation, 2025); Ian Ayres and Jack Balkin, ‘The Law of AI Is the Law of Risky Agents Without Intentions’ (2024) The University of Chicago Law Review Online, https://lawreview.uchicago.edu/online-archive/law-ai-law-risky-agents-without-intentions.

²⁶ R (Bridges) v Chief Constable of South Wales Police* [2020] EWCA Civ 1058; Judiciary of England and Wales, Artificial Intelligence (AI) Guidance for Judicial Office Holders (31 October 2025); Ayres and Balkin, ‘The Law of AI Is the Law of Risky Agents Without Intentions’ (2024).

²⁷ Wisconsin Supreme Court, State v Loomis (2016); State v Pickett, 466 N.J. Super. 270 (App. Div. 2021); Judiciary of England and Wales, Artificial Intelligence (AI) Guidance for Judicial Office Holders (31 October 2025).

²⁸ Judiciary of England and Wales, Artificial Intelligence (AI) Guidance for Judicial Office Holders (31 October 2025), replacing earlier guidance issued in December 2023 and April 2025; The Lady Chief Justice’s Report 2024, noting ongoing review of judicial AI guidance, https://www.judiciary.uk/wp-content/uploads/2024/11/24.147_JO_LCJ-Annual-Report-2024_v8_WEB.pdf.

²⁹ Home Office, Public attitudes to police use of facial recognition technology (4 December 2025), https://www.gov.uk/government/publications/public-attitudes-to-police-use-of-facial-recognition-technology/public-attitudes-to-police-use-of-facial-recognition-technology; Home Office, Police use of facial recognition: factsheet (4 December 2025).

³⁰ Regulation (EU) 2024/1689 (Artificial Intelligence Act); Home Office, Legal framework for using facial recognition in law enforcement (consultation, 2025); National Conference of State Legislatures, Artificial Intelligence in Law Enforcement: The Federal and State Landscape (2024).

³¹ Home Office, Legal framework for using facial recognition in law enforcement (consultation, 2025); R (Bridges) v Chief Constable of South Wales Police [2020] EWCA Civ 1058.

³² R (Bridges) v Chief Constable of South Wales Police [2020] EWCA Civ 1058; Royal Commission into the Robodebt Scheme, Report (2023); European Parliament, Artificial Intelligence Act: MEPs adopt landmark law (13 March 2024).

The rise of Decentralised Autonomous Organisations (DAOs) has changed how organisations coordinate in the digital economy, shifting some governance functions from hierarchical management to code-based and community driven processes.^{[1, 3]} The friction between borderless, code-based systems and territorially bounded, person centred legal regimes has driven the use of ‘legal wrappers’, legal entities that connect on-chain governance to established financial and legal systems.^{[1, 2]} As DAOs moved from experimental communities to organisations administering substantial assets, questions of legal recognition, contracting capacity and liability exposure became central for participants and investors.^{[1, 3]}

A DAO legal wrapper is a legal entity used by a decentralised organisation to operate within existing legal systems, providing separate legal personality, limited liability and a more defined tax position.^[2] It allows a DAO to interact with traditional financial and legal infrastructure while preserving on-chain governance, including the ability to hold property, enter contracts and limit members’ personal exposure.^{[2, 4]} The global position remains fragmented, with jurisdictions taking different approaches to innovation, supervision and consumer protection.^{[5, 6]}

The Crisis of the Unwrapped DAO: Partnership Liability and Judicial Characterisation

The core legal risk for an unwrapped DAO is that courts or regulators may characterise it as a general partnership or an unincorporated association.^{[4, 7]} Under common law and many civil law systems, an association of persons carrying on a business for profit may be treated as a partnership even if no formal entity was intended.^{[8, 9]} That characterisation can expose participants to joint and several liability, including personal responsibility for organisational debts, torts and regulatory breaches.^{[5, 10]}

Recent US litigation has accelerated this analysis.^{[11, 12]} Courts have shown little appetite for the argument that a DAO is merely ‘software’ and therefore outside person-based legal frameworks.^{[11, 12]} The practical lesson is straightforward: the absence of a legal wrapper does not eliminate legal risk and may increase it.^{[8, 10]}

CFTC v. Ooki DAO

CFTC v. Ooki DAO is a leading authority on a DAO’s capacity to be sued and on the exposure of governance participants.^{[10, 13]} The case arose after bZeroX, LLC transferred protocol control to the bZx DAO (later renamed Ooki DAO), in part to decentralise operations and, on the CFTC’s case, to continue regulated activity through a DAO structure.^{[8, 10]} The CFTC alleged that the DAO operated as an unregistered futures commission merchant and failed to implement required AML and KYC controls.^{[13, 14]}

Service of process was a central procedural issue.^{[11, 15]} Because Ooki DAO had no physical address or registered agent, the CFTC sought service through the DAO’s online forum and help chat box.^{[11, 15]} In earlier service proceedings (later referenced in the default judgment), the court accepted that method as reasonably calculated to provide notice, confirming that a DAO’s digital channels can be used for service in appropriate circumstances.^{[11, 14]} The court also held that Ooki DAO qualified as an unincorporated association under California law.^{[8, 11]} That finding materially increased personal liability risk for governance participants by supporting the theory that voting members may be treated as members of the association for regulatory purposes.^{[10, 15]}

Samuels v. Lido DAO and the Targeting of Governance Participants

The litigation in Samuels v. Lido DAO further develops the pleading stage framework for personal liability in DAO governance.^{[12, 16]} The plaintiff alleges that Lido DAO offered and sold unregistered securities through LDO tokens.^{[12, 16]} At the motion to dismiss stage, the court allowed key claims to proceed and accepted, for pleading purposes, the theory that Lido DAO could be treated as a general partnership under California law.^{[12, 16]} A notable aspect of the ruling is its focus on ‘meaningful participation’, with allegations directed at venture investors including Paradigm, Andreessen Horowitz and Dragonfly based on their governance activity.^{[9, 16]}

The court’s reasoning at that stage referred to factors such as governance rights, proposal activity, voting conduct and public statements suggesting active involvement.^[16] The result is a tiered litigation risk profile in which visible or active governance participants may face greater exposure than passive token holders.^[16] For venture investors, the case is a reminder that active DAO governance may be relied on by claimants as evidence of partnership style participation where no wrapper exists.^[16]

Table 1. Selected DAO litigation examples: allegations, legal characterisation and procedural status.

The Taxonomy of DAO Legal Wrappers

To mitigate these risks, DAOs have adopted a range of wrapper structures designed to provide separate legal personality and to separate organisational liabilities from members’ personal assets.^{[2, 4]} Selecting a wrapper is a strategic legal and operational decision that requires trade offs among cost, regulatory posture, tax treatment and the degree of centralisation needed for execution.^{[1, 18]}

Foundations and Ownerless Entities

Foundation structures, particularly in jurisdictions such as the Cayman Islands, the British Virgin Islands and Switzerland, are widely used for larger token communities.^{[18, 19]} These entities are often described as ‘ownerless’ in the sense that they do not have shareholders or members in the ordinary corporate form.^{[18, 20]} They are instead constituted to pursue a stated purpose, such as protocol stewardship, treasury administration or ecosystem development.^{[21, 22]}

The Cayman Islands Foundation Company (CFC) is often used because it can combine company style governance with purpose driven features.^{[19, 20]} Its constitutional documents can be drafted to recognise on-chain voting outcomes as part of the foundation’s decision making process.^{[19, 22]} Directors or supervisors then act as the execution layer for off-chain functions such as banking, contracting and IP ownership.^{[20, 21]} Used properly, this structure can materially reduce the risk of a general partnership characterisation by providing a clear legal counterparty and liability shield.^{[18, 19]}

Limited Liability Companies (LLCs) and Special Jurisdictions

The LLC model is often used for smaller or more closely held DAOs, including investment groups and service collectives.^[1] Although an LLC offers a familiar liability shield, it can introduce identifiable control points that may sit uneasily with a DAO’s decentralisation objectives.^[1] Wyoming and the Marshall Islands have both enacted DAO-specific or DAO enabling LLC legislation to bridge that gap.^{[5, 23, 24]}

In the Marshall Islands, the Decentralized Autonomous Organization Act 2022 recognises a DAO LLC as a resident domestic LLC that elects DAO status and links governance to its constitutional documents and, where applicable, smart contracts.^[23] The statute allows one or more members, requires a registered agent in the Republic and contemplates both for profit and non-profit DAO LLC structures.^[23] It also applies tailored disclosure and reporting rules, including thresholds for significant governance rights, and integrates DAO LLCs into the wider Marshall Islands LLC framework.^[23]

Unincorporated Nonprofit Associations (UNAs) and the DUNA Model

The UNA model takes a different route by seeking to avoid the ‘for profit business’ criterion often relied on in implied partnership analysis.^{[4, 9]} By organising as a nonprofit association, a DAO may argue that it lacks the co ownership of a business for profit that commonly underpins partnership findings under state law, including in California.^[9] Wyoming’s Decentralised Unincorporated Nonprofit Association (DUNA) Act 2024 is the most developed example of this approach.^[9]

The DUNA framework provides several strategic advantages:

• Separate Legal Personality: A DUNA is a legal entity separate from its members for contract and tort purposes, supporting a meaningful liability shield if the statutory requirements are met.^[9]

• Nonprofit Form with Permitted Activity: The statute permits revenue generating activity, provided it is pursued in furtherance of the association’s stated nonprofit purpose rather than as a profit distribution vehicle.^[9]

• Tax Structuring Flexibility: The framework may improve tax administration and classification planning for some DAOs, but tax outcomes remain fact-specific and require jurisdiction-specific advice rather than assumption.^[9]

• Securities Analysis (Not a Safe Harbour): A nonprofit form may assist arguments about token holder expectations and governance rights, but it does not remove securities law risk or displace the Howey analysis.^[9]

Table 2. Comparative DAO wrapper jurisdictions: indicative registration features and strategic benefits.

The European and Middle Eastern Regulatory Response

While the United States has tended toward a mix of state level wrapper legislation and federal enforcement activity, several European and Middle Eastern jurisdictions have focused on integrating DLT structures into established legal frameworks.^{[5, 6, 31]}

Switzerland: The Institutional Hub for DLT

Switzerland’s Federal Act on the Adaptation of Federal Legislation to Developments in Distributed Ledger Technology (the DLT Act), which was implemented in stages in 2021, is one of the most developed technology neutral frameworks for integrating DLT into existing federal law.^{[6, 27]} Rather than creating a parallel crypto code, it amends existing legislation across areas including private law, insolvency and financial market infrastructure.^{[6, 27]}

A key feature of the Swiss reforms is the recognition of ledger-based securities, allowing certain rights to be represented and transferred via distributed ledger systems with legal effect under Swiss law.^{[27, 28]} For DAOs, Switzerland commonly presents two practical vehicles: the Swiss Association and the Swiss Foundation.^[25] The association is generally more flexible and cost efficient for community led projects, while the foundation is more formal and supervised, and may be preferred where larger treasuries, grant programmes or institutional counterparties are involved.^{[25, 29]}

The UK Law Commission’s Approach

In 2026, the Law Commission of England and Wales published a scoping paper on DAOs and concluded that there was no immediate need for a DAO-specific legal entity in England and Wales.^[5] Instead, it recommended reviewing existing frameworks, including the Companies Act 2006 and LLP legislation, to assess how they can better accommodate technology enabled governance.^[5] The paper also identified a possible role for a limited liability not for profit association, indicating a cautious but practical reform path-based on adaptation of familiar legal forms rather than creation of a wholly new category.^[5]

ADGM and the DLT Foundation Framework

Abu Dhabi Global Market (ADGM) has moved beyond consultation and introduced a bespoke DLT Foundations regime through the Distributed Ledger Technology Foundations Regulations 2023.^[31] The framework is designed to provide separate legal personality and a structured registration and governance regime for DLT foundations, including use cases relevant to DAOs and token-based projects.^[31] As with other regulated wrappers, the benefits of legal certainty are paired with compliance obligations, including AML and related governance requirements that may not suit anonymous or permissionless community models.^[31]

Judicial Disputes in Protocol Liability and Market Manipulation

Beyond wrapper selection, recent disputes have tested the legal responsibility of developers, governance participants and traders in relation to protocol design, third party conduct and market manipulation theories.^{[17, 35]}

Uniswap Labs and the ‘Software Provider’ Defence

Risley v. Uniswap Labs is an important appellate decision on protocol developer liability, but it should not be overstated.^{[17, 33]} Private plaintiffs sought to hold Uniswap Labs and related investors responsible for losses on allegedly fraudulent third party tokens traded via the protocol.^[17] The Second Circuit affirmed dismissal of the federal claims, while vacating and remanding aspects of the state law claims for further proceedings.^{[17, 33]} The decision supports the argument that protocol developers are not automatically liable for all third party misuse of open software, while leaving scope for case specific claims depending on pleaded facts and legal theory.^{[33, 34]}

Mango Markets and the Eisenberg ‘Code is Law’ Conflict

Avraham Eisenberg’s exploitation of Mango Markets illustrates the tension between ‘code is law’ narratives and conventional fraud and market manipulation doctrine.^{[35, 36]} Prosecutors alleged that Eisenberg manipulated the price of MNGO related instruments and extracted approximately $110 million in crypto assets from Mango Markets.^[35] Although a jury convicted him in 2024, the SDNY later granted Rule 29 relief in part, vacating the commodities fraud and manipulation convictions on venue grounds and entering a judgment of acquittal on the wire fraud count for insufficient evidence.^[35] The decision shows that DeFi prosecutions still turn on orthodox criminal law requirements such as venue and proof of misrepresentation, even where the underlying platform is permissionless.^{[35, 36]}

Strategic Insights for Global Structuring

The global market for DAO wrappers is moving from informal experimentation toward more explicit legal structuring.^{[1, 18, 30]} Cases such as Ooki and Lido show that an unwrapped DAO can face ordinary partnership or association analysis, with potentially serious personal liability consequences for active participants.^{[10, 11, 12, 16]}

The Evolution of the ‘Orphan’ Entity

Purpose-based foundation structures in jurisdictions such as Cayman and Switzerland remain a common solution for large token communities.^{[1, 20]} By removing shareholder ownership and introducing a legal counterparty, these entities can preserve decentralised governance at the protocol layer while supporting contracting, treasury management and compliance at the legal layer.^{[19, 21]} They also introduce administrative burdens, including governance formalities, local service providers and, in some cases, KYC obligations, which may be difficult for highly decentralised communities to accept.^{[1, 29]}

Functional Equivalence and the Future of Regulation

The longer term regulatory question is whether legal systems will increasingly accept functional equivalence, as reflected in initiatives such as the COALA Model Law.^{[37, 38]} The underlying proposition is that some blockchain features, including transparency and immutability, may achieve policy objectives similar to parts of conventional corporate compliance frameworks.^{[37, 39]} If that approach gains wider acceptance, wrappers may become more streamlined, but they are unlikely to disappear in the near term given contracting, enforcement and jurisdictional realities.^{[37, 38]}

Tax Neutrality and International Private Law

As DAOs operate across borders, conflict of laws issues, including the lex situs of digital assets, remain difficult and commercially significant.^{[40, 41]} The UK Law Commission’s work on digital assets, including the proposal for a third category of personal property, is an important step toward greater legal certainty in cross border disputes and transactions.^[41] For now, many projects continue to use tax neutral or tax efficient jurisdictions for treasury and governance structures, while monitoring changes in international tax and regulatory standards.^{[19, 42]}

Synthesis and Conclusion

The current DAO legal landscape is developing along two tracks.^{[1, 30]} Courts and regulators continue to apply existing partnership, association and enforcement doctrines where no clear legal structure is present.^{[8, 10, 11, 13]} At the same time, some jurisdictions are building workable legal vehicles that allow DAO linked projects to operate with greater certainty.^{[23, 31]}

For advisers, jurisdiction selection is no longer a one-off branding exercise but a risk-based structuring decision tied to the DAO’s governance model, treasury profile, user base and operating footprint.^[1] A layered structure can be appropriate in some cases, for example a governance facing foundation combined with separate operating entities for specific activities or geographies.^[1] As legal frameworks mature, the wrapper is likely to remain a practical interface between decentralised governance and the legal system, even if its form becomes more specialised.^{[2, 3, 37]}

The move from informal online coordination to legally structured operation need not amount to abandonment of decentralisation.^{[1, 7]} For many projects, a wrapper is simply the mechanism that enables contracting, limited liability and regulatory engagement at scale.^{[2, 29]} The durability of DAO models will depend not only on protocol design, but also on the quality of their legal architecture.^{[1, 3]}

1. Coincub, “DAO Legal Wrappers Playbook.” https://coincub.com/blog/dao-legal-wrappers/

2. Legal Nodes, “DAO Legal Wrappers: Definition, Types, Jurisdictions and Use Cases.” https://legalnodes.com/article/dao-legal-wrapper

3. Law Commission of England and Wales, “Decentralised Autonomous Organisations (DAOs)” project page. https://lawcom.gov.uk/project/decentralised-autonomous-organisations-daos/

4. Legal Nodes, “Choose a Crypto-Friendly Country for a DAO.” https://legalnodes.com/article/choose-a-crypto-friendly-country-for-dao

5. Law Commission of England and Wales, “Decentralised Autonomous Organisations (DAOs): Scoping paper” (project publication page / paper). https://lawcom.gov.uk/project/decentralised-autonomous-organisations-daos/

6. Swiss State Secretariat for International Finance (SIF), “DLT / blockchain / tokenisation.” https://www.sif.admin.ch/en/dlt-blockchain-en

7. O’Melveny, “Decentralized Autonomous Organizations (DAOs): Overview.” https://www.omm.com/insights/alerts-publications/

8. Fenwick, “The Legal Landscape for DAOs: Key Lessons from Lido DAO and Ooki DAO.” https://www.fenwick.com/insights/publications/the-legal-landscape-for-daos-key-lessons-from-lido-dao-and-ooki-dao

9. Wyoming Legislature, Wyoming Decentralized Unincorporated Nonprofit Association Act (DUNA), Wyo. Stat. Title 17, Ch. 32. https://wyoleg.gov/Legislation/Statutes

10. Hodder Law, “CFTC and DAO Regulation / Ooki DAO case commentary.” https://hodder.law/cftc-ooki-dao-regulation/

11. Davis Wright Tremaine, Financial Services Law Advisor, DAO/CFTC litigation commentary (Ooki DAO). https://www.dwt.com/blogs/financial-services-law-advisor/2023/01/dao-cftc-digital-assets-blockchain-lawsuits

12. Davis Wright Tremaine, “Samuels v. Lido DAO: A Potential New Frontier for Liability in the Cryptocurrency Space.” https://www.dwt.com/blogs/financial-services-law-advisor/2025/01/lido-dao-crypto-liability-california-court-case

13. Commodity Futures Trading Commission, Press Release No. 8714-23 (Ooki DAO litigation / enforcement update). https://www.cftc.gov/PressRoom/PressReleases/8714-23

14. CourtListener docket, Commodity Futures Trading Commission v. Ooki DAO, No. 3:22-cv-05416 (N.D. Cal.). https://www.courtlistener.com/docket/65369411/commodity-futures-trading-commission-v-ooki-dao/

15. CourtListener docket entries and judgment materials, Commodity Futures Trading Commission v. Ooki DAO, No. 3:22-cv-05416 (N.D. Cal.). https://www.courtlistener.com/docket/65369411/commodity-futures-trading-commission-v-ooki-dao/

16. Practical Law (Thomson Reuters), Samuels v. Lido DAO case update (subscription source). https://uk.practicallaw.thomsonreuters.com/

17. Risley v. Universal Navigation Inc. (Uniswap), U.S. Court of Appeals for the Second Circuit (summary order / docket materials via CourtListener). https://storage.courtlistener.com/

18. Wyoming Legislature, Wyoming DAO Supplement (DAO LLC), Wyo. Stat. Title 17, Ch. 31. https://wyoleg.gov/Legislation/Statutes

19. Cayman Islands, Foundation Companies Act (as revised) (official legislation / official hosted text). https://www.cima.ky/upimages/commonfiles/FoundationCompaniesAct2017_1737729600.PDF

20. Collas Crill, “Cayman Islands foundation companies … Web3 innovation” (foundation company structuring commentary). https://www.collascrill.com/articles/

21. HCS Offshore Services Ltd., “Cayman Islands Foundation Companies and DAOs” (white paper).

22. Legal Nodes, “Caymanian Foundation for DAO.” https://www.legalnodes.com/article/caymanian-foundation-for-dao

23. Republic of the Marshall Islands, Decentralized Autonomous Organization Act 2022 (Public Law 2022-50). https://rmicourts.org/wp-content/uploads/2022/12/PL-2022-50-Decentralized-Autonomous.pdf

24. Stinson LLP, “Decentralized Autonomous Organization Laws Across the U.S.” https://www.stinson.com/newsroom-publications-DAO-laws-across-the-US

25. Vectra Advisors, “Understanding DAOs and Legal Wrappers in Switzerland.” https://www.vectraadvisors.com/blog/understanding-daos-and-legal-wrappers-in-switzerland

26. Pontinova Law, “Switzerland Foundation” / DAO structuring note. https://www.pontinova.law/dao/switzerland-foundation

27. Swiss Federal Council / Swiss federal authorities, DLT legislative framework information (incl. implementation news). https://www.news.admin.ch/en/nsb?id=84035

28. Swiss Code of Obligations (Fedlex; see Art. 973d et seq. on ledger-based securities) and related Fedlex materials. https://www.fedlex.admin.ch/eli/cc/27/317_321_377/en

29. DAOBox Docs, “Swiss Foundation as a DAO Legal Wrapper - Complete Guide.” https://docs.daobox.io/international/swiss-foundation-as-a-dao-legal-wrapper-complete-guide

30. Oxford University Press, Capital Markets Law Journal, “Decentralized autonomous organizations: adapting legal structures to fit new digital environments” (2025). https://academic.oup.com/cmlj/article/20/3/kmaf011/8249442

31. ADGM, Distributed Ledger Technology Foundations Regulations 2023 (official ADGM/Thomson Reuters rulebook page). https://en.adgm.thomsonreuters.com/rulebook/distributed-ledger-technology-foundations-regulations-2023

32. Uniswap Labs blog, “A Win for DeFi” (SEC investigation closure announcement). https://blog.uniswap.org/a-win-for-defi

33. Holland & Knight / media coverage note on Second Circuit Uniswap-related ruling. https://www.hklaw.com/en/news/intheheadlines/

34. Columbia Law School Blue Sky Blog, “Uniswap’s Reprieve Reveals the Uncertainty of DeFi Regulation.” https://clsbluesky.law.columbia.edu/2025/03/13/uniswaps-reprieve-reveals-the-uncertainty-of-defi-regulation/

35. United States v. Avraham Eisenberg, No. 23-cr-10 (S.D.N.Y.), Opinion and Order (23 May 2025). https://nysd.uscourts.gov/sites/default/files/2025-05/23cr10%20Opinion%20and%20Order.pdf

36. TRM Labs, commentary on the Eisenberg / Mango Markets decision. https://www.trmlabs.com/post/

37. COALA, “The DAO Model Law” (Medium publication). https://medium.com/coala/the-dao-model-law-68e5360971ea

38. UNIDROIT, Digital Assets and Private Law project materials (institutional comparator; not a DAO-specific model law). https://www.unidroit.org/work-in-progress/digital-assets-and-private-law/

39. UNCITRAL / UN General Assembly working document A/CN.9/1225 (DAO-related/private law digital issues reference). https://docs.un.org/en/A/CN.9/1225

40. Law Commission of England and Wales, “Digital assets and electronic trade documents in private international law” project page. https://lawcom.gov.uk/project/digital-assets-and-electronic-trade-documents-in-private-international-law/

41. Law Commission of England and Wales, “Digital assets” project page. https://lawcom.gov.uk/project/digital-assets/

42. Bedell Cristin, “Foundation Companies as DAOs.” https://www.bedellcristin.com/knowledge/briefings/fy-2223/foundation-companies-as-daos/

The construction industry has historically operated as a high-stakes, low-margin sector where technological adoption follows a trajectory of cautious pragmatism. However, the period between 2021 and 2026 represents a departure from this traditional path, as the integration of Artificial Intelligence (AI) and 3D printing (3DP) has fundamentally altered the risk profiles of major infrastructure and residential projects globally. This transformation is not merely operational; it has permeated the legal foundations of the industry, giving rise to new forms of litigation and arbitration that challenge established doctrines of professional negligence, design liability, and procedural due process. As project data volumes explode and autonomous systems begin to occupy roles once held by licensed professionals, the legal community is grappling with the “liability squeeze” that occurs when traditional contracts fail to account for the nuances of machine-generated errors and the complexities of additive manufacturing.^{[1, 2]}

The Emergence of Procedural AI in Construction Disputes

The first significant wave of litigation and arbitration involving AI in construction has not originated from structural collapses caused by algorithms, but from the procedural use of AI within the dispute resolution process itself. Construction disputes are notoriously data-heavy, often involving millions of project emails, complex schedules and thousands of change orders.^{[3, 4]} In the last five years, AI has transitioned from a backend tool for e-discovery to a front-facing participant in the adjudication of claims. The year 2025 is noted as the first full year where applied generative AI became a standard fixture in handling complex construction dispute cases, enabling legal teams to process 100,000 emails in under a week - a task that previously required seven weeks and hundreds of thousands of dollars in manual legal labour.^{[4, 5]}

However, this efficiency has introduced a new class of litigation: sanctions for the misuse of AI in court filings. The case of Noland v. Land of the Free, L.P. illustrates the growing judicial intolerance for “AI hallucinations,” where a lawyer was sanctioned $10,000 for filing appellate briefs containing fabricated case citations generated by ChatGPT.^[6] Such incidents have prompted a global regulatory response, with over 39 federal judges in the United States and various international bodies, such as the Dubai International Arbitration Centre (DIAC), issuing standing orders that require the disclosure of AI use and rigorous human verification of all AI-generated content.^{[6, 7]}

The launch of the AAA-ICDR AI-based arbitrator in November 2025 represents a paradigm shift in construction arbitration. Trained on more than 1,500 construction awards, the system delivers draft awards for low-to-mid-value, two-party disputes.^{[6, 10]} While this “human-in-the-loop” model maintains a requirement for a human arbitrator to validate and sign the final award, it has raised profound questions regarding the enforceability of such decisions under the New York Convention. Critics argue that the delegation of decisional mandates to an algorithm could be viewed as arbitrator misconduct or a violation of the tribunal’s proper constitution, potentially leading to the vacatur of awards in jurisdictions that require human adjudication.^[10]

Substantive Liability and the Professional Standard of Care

As AI moves beyond procedural support and into the realm of design and project management, the industry is witnessing the emergence of substantive liability claims. These disputes centre on the question of who bears responsibility when an AI-driven tool produces a flawed output that results in physical failure or financial loss. Traditional construction law defines the standard of care for architects and engineers as the “reasonable care and competence” ordinarily exercised by peers in good standing in the same locality.^[11] However, the proliferation of AI-enabled design and monitoring tools is rapidly blurring the line between this simple peer comparison and the “state of the art” standard found in product liability law.^[11]

The Shift from Reasonable Care to State of the Art

In 2025 and 2026, the gap between what is technically feasible and what is required for “reasonable competence” has begun to close. A design professional who refuses to utilise AI-powered tools for clash detection or structural optimisation may soon be viewed as ignoring practical tools that improve safety, potentially exposing them to negligence claims.^[11]

Conversely, the “unverified reliance” on AI outputs is becoming a primary driver of professional indemnity claims.^[1] If an architect follows a generative design system’s proposal for a structurally unsound layout in good faith, the liability will likely still reside with the professional for failing to exercise the requisite oversight, rather than with the software provider, who often disclaims all warranties through restrictive licensing agreements.^{[1, 12]}

The legal implications are particularly severe in projects utilising Building Information Modelling (BIM) and digital twins. BIM functions as a “digital rehearsal” of the construction process, but when protocols for data ownership and modification are unclear, disputes erupt over “blurred lines of responsibility”.^{[13, 16]} In the US, the Spearin Doctrine, which traditionally implies a warranty of design adequacy from the owner to the contractor, may be compromised in a collaborative BIM environment where contractors and suppliers contribute data to a shared model.^{[16, 17]} In such cases, the contractor might lose the protection of the Spearin Doctrine if they are seen as co-creators of the flawed digital design.^[16]

3D Printing in Construction: Case Studies in Failure and Litigation

The application of 3D printing in construction (3DCP) has moved from experimental pilots to commercial reality, but it has not been without significant technical and legal setbacks. The last five years have seen the first major instances of structural failure and subsequent litigation or arbitration related to additive manufacturing.

The Iowa Demolition: A Case Study in PSI Deficiency

In 2023, a project in Muscatine, Iowa, intended to produce ten 3D-printed homes, faced a major failure when the first completed structure failed to meet the required compressive strength standards.^{[18, 19]} Although the material mixture, which included hempcrete, met laboratory requirements, the on-site execution resulted in a failure consistently to reach the minimum threshold needed for structural safety.^{[18, 19]}

This incident highlights several emerging litigation themes in 3DCP:

1. Material Performance Discrepancies: The disconnect between laboratory testing and on-site performance of proprietary cementitious mixtures creates significant risk for contractors and material suppliers.^[19]

2. Contractual Absorption of Costs: In the Iowa case, the subcontractor, Alquist 3D, absorbed the costs of the demolition and reconstruction.^[19] This “private resolution” is common in the current stage of the industry, as pioneers seek to avoid public litigation that might damage the reputation of the technology.

3. Mechanical and Environmental Delays: 3D printing is sensitive to weather conditions; delays can cause problems with “interlayer bonding” if one layer hardens before the next is laid.^{[17, 20]} This “delamination” is a frequent subject of arbitration claims regarding long-term structural durability.^[17]

Arbitration and Structural Performance Failures

Beyond total demolitions, arbitration has become the preferred mechanism for resolving structural integrity issues in US 3D-printed housing projects. Common disputes involve “load-bearing or seismic-resistance failures” and “non-compliance with local building codes”.^[17] Arbitrators in these cases are increasingly relying on traditional principles, such as the economic-loss doctrine and the Spearin Doctrine, but adapting them to “hybrid responsibility models” where designers, software providers, and robotics operators all play a role in the construction process.^[17]

Intellectual Property Litigation in the 3D Printing Sector

The rise of 3D printing has also triggered a surge in intellectual property (IP) litigation, particularly between established hardware manufacturers and newer market entrants. The rapid evolution of the technology has outpaced the ability of some companies to secure traditional patent protection, leading to claims of design theft and unauthorised replication.

The Stratasys v. Bambu Lab Conflict

In 2024, the US-based company Stratasys filed a landmark claim against China-based Bambu Lab, alleging the infringement of several patents related to 3D printing hardware.^[21] The patents in question - covering purge towers, heated build platforms and sensor-based bed mapping - are fundamental to the accuracy and efficiency required for large-scale construction printing.^[21] This case underscores the “IP theft window,” where the average 2.4-year duration of patent litigation allows a competitor to replicate a design and refine it before a trial can even begin.^[21]

The democratisation of manufacturing through 3DP has also raised concerns regarding the “right to repair.” In the EU, directives on spare parts are impacting how companies protect their IP for 3D-printed components, potentially allowing for the unauthorised printing of spare parts for construction machinery under certain conditions.^{[25, 27]} This creates a tension between patent holders and third-party 3DP services that manufacturers must navigate through complex licensing models.^[27]

Global Regulatory Trends: EU, China, and the US (2025–2026)

The regulatory landscape for AI and 3DP in construction is currently undergoing a massive transformation, with major frameworks coming into force.

The EU AI Act and High-Risk Infrastructure

The European Union’s AI Act, which became law in late 2024, classifies AI used in “critical infrastructure” as high-risk.^{[30, 31]} This has immediate implications for construction projects involving energy grids, transportation networks, and public buildings.

• Prohibited Systems (February 2025): Systems that manipulate human behaviour or exploit vulnerabilities are strictly banned.^{[32, 33]} This could apply to AI-driven labour monitoring tools that use subliminal techniques to influence worker productivity on jobsites.^[32]

• High-Risk Obligations (August 2026): Providers of AI systems for critical infrastructure must conduct mandatory “conformity assessments,” implement robust human oversight, and ensure the use of high-quality datasets to prevent algorithmic bias.^{[30, 34, 35]}

China’s “AI Plus” and Cybersecurity Frameworks

China has adopted a state-led approach to AI integration, through the “AI Plus Action Plan” issued in August 2025.^[36] The goal is to achieve 70% AI penetration in key industrial sectors, including construction and manufacturing, by 2027.^[36] However, this is coupled with rigorous cybersecurity amendments, effective January 1, 2026, which expand the state’s reach to “extraterritorial” cyber activities that endanger China’s network security.^{[37, 38]} For international firms, this means that failures in AI governance on projects in China could lead not only to civil litigation but to significant administrative fines and “app closure powers” by regulators.^[37]

US National Defence and Procurement Restrictions

In the United States, the 2026 National Defense Authorization Act (NDAA) has introduced a major shift by treating 3D printing as “strategically sensitive infrastructure”.^{[39, 40]} Section 849 of the Act prohibits the Department of Defense from procuring 3D printers from companies linked to China, Russia, Iran, or North Korea.^[40] This national security-based vetting process is reshaping the competitive landscape for construction technology, forcing defence contractors to audit their entire supply chains for “covered additive manufacturing machines”.^{[39, 40]}

Contractual Evolution: FIDIC, NEC4, and AI-Specific Clauses

Standard form contracts, which are the backbone of the international construction industry, are currently being rewritten to handle the risks of AI and 3DP. The consensus among drafters is that “contracts don’t deliver projects; people do,” but the presence of AI as a decision-maker requires explicit contractual recognition.^{[1, 41]}

NEC4 and the Spirit of Collaboration

The NEC4 suite of contracts, particularly in jurisdictions like Hong Kong and the UK, is being updated with Z-clauses that focus on “disclosure obligations” and “data ownership”.^{[12, 42]}

• Disclosure of AI Use: Contractors are increasingly required to reveal whether, where, and how AI is used in the delivery of services.^[12]

• Data Integrity Warranties: New clauses include warranties that AI systems have been trained on “representative and appropriately curated datasets” to guard against biased outcomes.^[43]

• Early Warning Systems: The NEC4 “early warning” mechanism is being leveraged to report potential AI malfunctions or “model drift” before they lead to project defects.^{[43, 44]}

FIDIC and the Future of the “Engineer”

The FIDIC 2025 conference in London highlighted the “Engineers’ accountability” in the age of AI.^{[41, 45]} A central debate is whether FIDIC should adopt AI into its multi-tiered dispute resolution mechanism. While some advocate for an “AI tier” before the Dispute Adjudication and Avoidance Board (DAAB) stage, others remain wary of the “black box” nature of algorithmic determinations.^{[41, 46]} FIDIC is currently urging the industry to include basic principles in contracts to begin measuring and monitoring AI performance, even without the immediate introduction of liquidated damages for AI failure.^{[41, 45]}

Insurance Market Reactions: Exclusions and “Silent AI”

The insurance industry is currently in a state of “racing to define its risk appetite” for AI in construction.^[47] This has led to two distinct trends: the introduction of “absolute” exclusions and the emergence of “Silent AI” risks.

The Proliferation of AI Exclusions

Between 2024 and 2026, major carriers like Berkley and Hamilton Insurance Group introduced broad exclusions for professional liability (PI) and directors and officers (D&O) policies.^{[47, 48]}

• The Berkley Exclusion: An “absolute” exclusion for any claim based upon or arising out of the actual or alleged use, deployment, or development of AI.^[47]

• Generative AI Exclusions: Specifically targeting the use of tools like ChatGPT or Midjourney in architectural design, these exclusions remove coverage for any damages or defence costs associated with content generated by these systems.^[47]

The Danger of Silent AI

“Silent AI” refers to risks that are neither explicitly included nor excluded in a policy, creating massive ambiguity and potential coverage gaps.^[49] For construction firms, this means a structural failure caused by a 3D printer malfunction might be denied under a standard PI policy because it involved an “autonomous system,” but also denied under a property policy due to cyber-related exclusions.^{[49, 50]} The market is currently seeing a small handful of AI-specific products, such as those from Armilla AI, but these rarely cover “bodily injury or property damage,” leaving the most critical construction risks unaddressed.^[51]

Conclusion: Strategic Implications for the Global Construction Sector

The evidence from the last five years indicates that the construction industry has entered a “liability squeeze” where the efficiency gains of AI and 3DP are offset by the emergence of novel legal and financial risks. Litigation and arbitration are no longer merely reactive processes but are becoming data-centric environments where the “human-in-the-loop” is a legal requirement for the enforceability of awards and the defence of professional standards.^{[10, 52]}

For contractors and developers, the key takeaways for 2026 and beyond are:

1. Contractual Clarity is Paramount: Relying on standard SaaS or “off-the-shelf” technology agreements is a significant liability. Contracts must explicitly allocate responsibility for AI inaccuracies and define the ownership of “derivative data” generated by project models.^{[1, 53]}

2. The “Spearin” Risk in BIM: The collaborative nature of modern design-build projects using AI-integrated BIM could deprive contractors of traditional defences against design errors. Detailed BIM protocols that define data contribution and approval stages are essential.^[16]

3. Vigilance in Insurance Renewals: As carriers introduce broad AI exclusions, firms must engage with brokers to identify “Silent AI” gaps and seek specialised endorsements where necessary.^{[49, 50]}

4. Regulatory Compliance as a Global Strategy: With the EU AI Act and China’s Cybersecurity amendments taking full effect in 2026, international construction firms must adopt a “highest common denominator” approach to AI governance to avoid catastrophic fines.^{[30, 33, 37]}

The “Year of Generative AI” in construction has arrived, but it has brought with it a complex web of litigation, arbitration and regulatory hurdles that will take decades to fully resolve. The industry’s success will depend not on the speed of its machines, but on the sophistication of its legal and risk management frameworks in navigating this unprecedented technological frontier.

--------------------------------------------------------------------------------

1. Who’s Liable When AI Builds? - Area Development, https://www.areadevelopment.com/construction-project-planning/q4-2025/whos-liable-when-ai-builds.shtml

2. AI Vendor Liability Squeeze: Courts Expand Accountability While Contracts Shift Risk, https://www.joneswalker.com/en/insights/blogs/ai-law-blog/ai-vendor-liability-squeeze-courts-expand-accountability-while-contracts-shift-r.html?id=102l4ta

3. How AI Is Changing Construction Disputes at AAA, https://www.adr.org/news-and-insights/ai-in-construction-disputes/

4. Construction Disputes in 2025: The Year of Generative AI | ThinkSet - BRG, https://www.thinkbrg.com/thinkset/construction-disputes-in-2025-the-year-of-generative-ai/

5. Construction Disputes in 2025: The Year of Generative AI is Here, https://www.americanbar.org/groups/construction_industry/resources/under-construction/2026-midwinter/construction-disputes-in-2025-the-year-of-generative-ai-is-here/

6. Courts and governments are still grappling with position of AI in ..., https://www.pinsentmasons.com/out-law/analysis/ai-litigationarbitration-legal-challenges

7. Arbitration group unveils AI tool for construction, https://www.constructiondive.com/news/arbitration-group-unveils-ai-tool-for-construction/805909/

8. Arbitration and AI: From Data Processing to Deepfakes. Outlining the Potential – and Pitfalls, https://www.klgates.com/Arbitration-and-AI-From-Data-Processing-to-Deepfakes-Outlining-the-Potential-and-Pitfalls-of-AI-in-Arbitration-10-27-2025

9. Professional Liability Risks in the Age of Artificial Intelligence | DWF Group, https://dwfgroup.com/en/news-and-insights/insights/2025/9/professional-liability-risks-in-the-age-of-artificial-intelligence

10. International Arbitration Experts Discuss The Use Of An AI Arbitrator ..., https://www.alvarezandmarsal.com/thought-leadership/international-arbitration-experts-discuss-the-use-of-an-ai-arbitrator-for-construction-arbitrations

11. Navigating the AI revolution: Duty of care in modern design | News & Events | Clark Hill PLC, https://www.clarkhill.com/news-events/news/navigating-the-ai-revolution-duty-of-care-in-modern-design/

12. The AI Contract Conundrum Beyond Standard Terms - Bird & Bird, https://www.twobirds.com/en/insights/2025/the-ai-contract-conundrum-beyond-standard-terms

13. Digital Construction: BIM, Smart Contracts, and AI Challenges - Youssef & Partners, https://youssef.law/insights/digital-construction-bim-smart-contracts-ai-challenges/

14. How is AI transforming international construction projects and what legal disputes could follow? - Gowling WLG, https://gowlingwlg.com/en/insights-resources/articles/2025/how-is-ai-transforming-international-construction-projects

15. The Legal Risks of AI in the Homebuilding Industry | Tech Talk | Insights & Events, https://www.bilzin.com/insights/publications/2025/08/the-legal-risks-of-ai-in-homebuilding

16. Legal Considerations Associated with Building Information Modeling, https://caed.calpoly.edu/content/pdci/research-projects/simonian-10

17. Arbitration Involving Disagreements Regarding 3D ... - Law Gratis, https://www.lawgratis.com/blog-detail/arbitration-involving-disagreements-regarding-3d-printed-housing-structural-integrity-issues-in-us-construction-projects

18. Iowa Demolishes Its First 3D Printed House - Hackaday, https://hackaday.com/2023/11/29/iowa-demolishes-its-first-3d-printed-house/

19. Why did Iowa demolish its first 3D printed house? - 3D ADEPT MEDIA, https://3dadept.com/why-did-iowa-demolish-its-first-3d-printed-house/

20. 3D Construction Trends And 3D Concrete Printing | AJG United States - Gallagher Insurance, https://www.ajg.com/news-and-insights/3d-construction-trends-and-3d-concrete-printing/

21. China Is Winning the 3D-Printing Race. Can the US Catch Up? | PCMag, https://www.pcmag.com/opinions/china-is-winning-the-3d-printing-race-can-the-us-catch-back-up

22. 3D printing and IP law - WIPO, https://www.wipo.int/en/web/wipo-magazine/articles/3d-printing-and-ip-law-39896

23. 3D printing and housing: intellectual property and construction law - UQ eSpace, https://espace.library.uq.edu.au/view/UQ:f762d90

24. Intellectual Property Challenges in the Age of 3D Printing: Navigating the Digital Copycat Dilemma - MDPI, https://www.mdpi.com/2076-3417/14/23/11448

25. 3D Printing Technology and Intellectual Property Infringement - Myerson Solicitors, https://www.myerson.co.uk/news-insights-and-events/3d-printing-technology-and-intellectual-property-infringement

26. Implications of 3D Printing in Intellectual Property, https://abounaja.com/blog/3d-printing-and-intellectual-property

27. 3D printing and intellectual property: Essential insights for IP professionals, https://ipbusinessacademy.org/3d-printing-and-intellectual-property-essential-insights-for-ip-professionals

28. Legal challenges of 3D printing | DLA Piper, https://www.dlapiper.com/insights/publications/2015/05/law-a-la-mode-issue-16/legal-challenges-of-3d-printing

29. 3D printing technology and intellectual property infringement - BusinessCloud, https://businesscloud.co.uk/opinion/3d-printing-technology-and-intellectual-property-infringement/

30. Key insights into AI regulations in the EU and the US: navigating the evolving landscape, https://www.kennedyslaw.com/en/thought-leadership/article/2025/key-insights-into-ai-regulations-in-the-eu-and-the-us-navigating-the-evolving-landscape/

31. AI Regulations in 2025: US, EU, UK, Japan, China & More - Anecdotes AI, https://www.anecdotes.ai/learn/ai-regulations-in-2025-us-eu-uk-japan-china-and-more

32. EU AI Act - Updates, Compliance, Training,

https://www.artificial-intelligence-act.com/

33. The First Requirements of the EU AI Act Come into Force in February 2025 | Littler, https://www.littler.com/news-analysis/asap/first-requirements-eu-ai-act-come-force-february-2025

34. High-level summary of the AI Act | EU Artificial Intelligence Act, https://artificialintelligenceact.eu/high-level-summary/

35. The Year in AI Law: 2025’s Biggest Legal Cases and What They Mean for 2026, https://www.internetlawyer-blog.com/the-year-in-ai-law-2025s-biggest-legal-cases-and-what-they-mean-for-2026/

36. Global AI Governance Law and Policy: China - IAPP, https://iapp.org/resources/article/global-ai-governance-china

37. China’s 2025 Cybersecurity Law amendments: Enhanced penalties, expanded extraterritorial application, and AI governance - Linklaters - Tech Insights, https://techinsights.linklaters.com/post/102lrz5/chinas-2025-cybersecurity-law-amendments-enhanced-penalties-expanded-extraterr

38. China Updates Cybersecurity Frameworks to Address AI and Infrastructure Risks, https://www.pearlcohen.com/china-updates-cybersecurity-frameworks-to-address-ai-and-infrastructure-risks/

39. Not a China Ban, but Close: Who the U.S. Military Can (and Can’t) Buy 3D Printers From, https://all3dp.com/4/not-a-china-ban-but-close-who-the-u-s-military-can-and-cant-buy-3d-printers-from/

40. U.S. blocks additive manufacturing systems tied to China, Russia, Iran, and North Korea from defense procurement - 3D Printing Industry, https://3dprintingindustry.com/news/u-s-blocks-additive-manufacturing-systems-tied-to-china-russia-iran-and-north-korea-from-defense-procurement-247868/

41. FIDIC CONTRACTS- HUMAN BEHAVIOURS, AI AND CARBON ..., https://www.bclplaw.com/en-US/events-insights-news/fidic-contracts-human-behaviours-ai-and-carbon-management.html

42. Embracing the NEC spirit: aligning Hong Kong practices for collaborative success | News, https://www.neccontract.com/news/embracing-the-nec-spirit-aligning-hong-kong-practices-for-collaborative-success

43. Eight essential clauses for AI contracts: A guide for vendors and customers in Northern Ireland - A&L Goodbody, https://www.algoodbody.com/insights-publications/eight-essential-clauses-for-ai-contracts-a-guide-for-vendors-and-customers-in-northern-ireland

44. Defects under NEC4 Contracts – summary of key terminology and mechanisms, https://www.sharpepritchard.co.uk/latest-news/defects-under-nec4-contracts-summary-of-key-terminology-and-mechanisms/

45. FIDIC Contracts - Human Behaviours, AI and Carbon Management | BCLP - JDSupra, https://www.jdsupra.com/legalnews/fidic-contracts-human-behaviours-ai-and-6054441/

46. Collaboration, AI, disputes and arbitration all highlighted at contracts conference - FIDIC, https://fidic.org/node/46434

47. AI Update: The Growing Trend of AI-Related Insurance Policy Exclusions | JD Supra, https://www.jdsupra.com/legalnews/ai-update-the-growing-trend-of-ai-3672112/

48. The Continued Proliferation of AI Exclusions - Hunton Andrews Kurth LLP, https://www.hunton.com/hunton-insurance-recovery-blog/the-continued-proliferation-of-ai-exclusions

49. Silent AI cover: the unforeseen risks for insurers - Kennedys Law, https://www.kennedyslaw.com/en/thought-leadership/article/2025/silent-ai-cover-the-unforeseen-risks-for-insurers/

50. Insuring the AI age - WTW, https://www.wtwco.com/en-us/insights/2025/12/insuring-the-ai-age

51. AI exclusions are creeping into insurance - but cyber policies aren’t the issue yet, https://www.insurancebusinessmag.com/us/news/cyber/ai-exclusions-are-creeping-into-insurance--but-cyber-policies-arent-the-issue-yet-560647.aspx

52. Construction professionals: how AI use can impact your PII - Lockton, https://global.lockton.com/gb/en/news-insights/construction-professionals-how-ai-use-can-impact-your-pii

53. The Rise of AI Vendor Agreements: 7 Clauses Every Business Needs to Get Right in 2025, https://holonlaw.com/ai/the-rise-of-ai-vendor-agreements/

The collapse of the FTX cryptocurrency exchange in November 2022 represented a definitive threshold in the history of financial technology, marking the transition from an era of unbridled speculative growth to one of institutional accountability and legislative maturity. What began as a liquidity crisis triggered by a series of investigative revelations and adversarial market manoeuvres rapidly unravelled into the exposure of one of the most sophisticated financial frauds in modern history.^{[1, 2, 3]}

This review provides a deconstruction of the mechanisms that led to the exchange’s failure, the ongoing efforts to remunerate victims, the judicial outcomes for the principal actors and the systemic lessons that have reshaped the regulatory landscape as of January 2026.

Historical Genesis: The Convergence of Arbitrage and Ambition

The foundations of the FTX collapse were laid years prior to its public insolvency, rooted in a structural lack of corporate boundaries between the exchange and its sister trading firm, Alameda Research. Founded in 2017 by Samuel Bankman-Fried and Tara MacAulay, Alameda Research initially presented itself as a high-frequency trading firm specialising in crypto-arbitrage.^{[1, 2]} Bankman-Fried, an adherent of the “Effective Altruism” movement, marketed the firm as a vehicle to generate wealth for the purpose of global philanthropic impact. However, the internal reality was characterised by high-risk strategies and a fundamental disregard for traditional risk management protocols.

As early as September 2018, Alameda was raising debt from investors by offering a 15% annualised fixed rate loan, promising “high returns with no risk”.^[1] This promise was inherently contradictory to the volatile nature of the cryptocurrency markets and suggested an early reliance on continuous capital inflows to sustain operations. When FTX was launched in November 2019, it was marketed as an exchange “built by traders, for traders.” The platform’s rapid ascent was fuelled by its user-friendly interface and the introduction of complex products like tokenised stocks and crypto derivatives.^{[2, 4]}

The strategic investment by Binance in late 2019 provided FTX with both capital and market credibility. However, the relationship between the two giants would eventually sour, leading to a buy-back of Binance’s equity stake in 2021, a transaction largely funded by the FTT token.^[6] This decision inadvertently created the ammunition that would eventually be used to trigger the exchange’s downfall.

The Mechanics of Malfeasance: Backdoors and the FTT Flywheel

The architectural instability of the FTX-Alameda nexus was predicated on a technological backdoor that bypassed the exchange’s automated risk mitigation systems. Testimony from senior executives like Gary Wang and Nishad Singh revealed that, at the direction of Bankman-Fried, the FTX software code was altered to grant Alameda Research a “virtually unlimited” line of credit.^{[3, 7]} While ordinary customers were subject to immediate liquidation if their collateral fell below required levels, Alameda was exempt from these controls.

The Duality of the FTT Token

A central component of this fragile ecosystem was the FTX Token (FTT). FTT was utilised not merely as a utility token for fee discounts but as a primary form of collateral that allowed Alameda to borrow against its own sister company’s perceived value.^{[1, 8]} By late 2021, Alameda’s balance sheet was heavily concentrated in FTT, creating a circular dependency where the solvency of the trading firm was inextricably linked to the market price of the exchange’s native token.

Alameda Research moved away from delta-neutral strategies in December 2021, beginning to take large directional risks on leverage.^[1] This shift coincided with the “crypto winter” of early 2022. When the Terra-Luna ecosystem collapsed in May 2022, it triggered a massive credit crunch. To hide the resultant losses at Alameda, FTX began lending the firm billions of dollars in customer funds - a decision Bankman-Fried later described to colleagues as a “poor judgment call”.^{[1, 6]}

The Shadow Account: North Dimension

The misappropriation was further obfuscated through the use of an entity called North Dimension. Before FTX established its own banking relationships, customers were directed to wire fiat currency to Alameda Research’s bank accounts. These funds were never consistently transferred to FTX. Instead, they were held by Alameda and used for its own operations, investments, and expenses.^{[7, 9]} Bankman-Fried contended in his defence that he believed these funds were being held as a “borrow,” but the lack of formal accounting meant that over $8 billion in customer liabilities were effectively invisible to the exchange’s internal dashboards.^{[9, 10]}

The Spiral of 2022: Contagion and Concealment

Throughout the summer of 2022, Bankman-Fried sought to project an image of financial strength by offering bailouts to struggling crypto platforms like Voyager Digital, Celsius and BlockFi.^[1] In reality, these actions were defensive manoeuvres intended to prevent the liquidation of Alameda’s positions on those platforms, which would have exposed its insolvency. By June 2022, Alameda had become the largest source of liquidity for FTX, accounting for a staggering 30% of USD Coin transfers and 10% of Tether transfers on the exchange.^[6]

The internal culture at the FTX headquarters in the Bahamas was characterised by an extreme lack of professional oversight. The executive team, which included Caroline Ellison as the sole CEO of Alameda after Sam Trabucco’s departure in August 2022, operated without a traditional management team or a functioning board of directors.^{[1, 11]} This vacuum allowed for the unchecked execution of high-risk strategies, including the use of customer funds for personal real estate purchases and millions of dollars in political contributions to candidates from both parties.^{[3, 4]}

The Seven Days in November: A Forensic Timeline

The terminal phase of the FTX collapse was initiated by a public loss of confidence. On November 2, 2022, CoinDesk published an article detailing Alameda’s balance sheet, which revealed that $6 billion of its $14.6 billion in assets were FTT tokens.^{[6, 8]} This exposure of the “circular collateral” model signalled that the firm’s equity was built on a foundation of its own creation.

The announcement by Changpeng Zhao (CZ) that Binance would sell its $529 million FTT position triggered an immediate 80% drop in the token’s value.^[6] Caroline Ellison attempted to stabilise the price by offering to buy Binance’s holdings at $22, which only intensified speculation that Alameda had loans that would be liquidated if the price fell below that threshold.^{[1, 8]} Unable to meet a $6 billion withdrawal demand, Bankman-Fried sought a rescue package of up to $9.4 billion from various investors before finally resigning and filing for bankruptcy.^{[1, 6]}

The Post-Collapse Void: The Administration of John J. Ray III

Upon the bankruptcy filing, John J. Ray III was appointed CEO. Ray’s first reports to the Delaware Bankruptcy Court were scathing, noting that “never in my career have I seen such a complete failure of corporate controls and such a complete absence of trustworthy financial information”.^{[2, 12]} His findings highlighted that FTX had no accounting department and that the function had been entirely outsourced to firms that lacked the capacity to manage a multi-billion dollar enterprise.^[11]

Control Failures and Security Breaches

The forensic analysis of the FTX infrastructure revealed systemic vulnerabilities. Key wallet secrets and operational credentials were stored in a shared cloud environment (including a single AWS account) creating a single point of compromise; passwords and private keys/seed phrases were poorly segregated and, in some instances, stored in plain text or committed to code repositories.^[13] Within roughly 24 hours of the Chapter 11 filing, FTX reported “unauthorised transactions,” and blockchain analytics firm Elliptic estimated that approximately $477 million in cryptoassets was moved out of FTX-controlled wallets.^[54]

Ray’s team discovered that the group’s financial records were so poor that they could not initially produce an accurate list of employees or bank accounts.^{[11, 12]} The consolidated assets as of September 30, 2022, were reported at approximately $13.4 billion, but this was heavily skewed by custodial liabilities and crypto asset borrowings that were not properly reflected in the books under Bankman-Fried’s control.^[5]

Judicial Reckoning: Criminal and Civil Outcomes

The legal proceedings against the FTX executive team concluded with significant prison sentences and permanent industry bans. As of January 2026, the principal architects of the scheme have seen their initial sentences upheld or reached the terminal phases of their incarceration.

Samuel Bankman-Fried: The Mastermind Narrative

On March 28, 2024, Samuel Bankman-Fried was sentenced to 25 years in federal prison and ordered to forfeit $11 billion.^{[3, 14]} Judge Lewis Kaplan summarily rejected the defence’s argument that the harm to customers was “zero” because the bankruptcy estate hoped to pay them back. The court characterised the fraud as one of the largest in history, second only to the prosecution of Bernie Madoff.^[10]

In November 2025, Bankman-Fried’s appeal was argued before the Second Circuit. His counsel contended that the trial court’s evidentiary rulings prevented the jury from hearing material evidence relating to counsel involvement and his asserted good faith beliefs; the appellate panel expressed scepticism, because Bankman-Fried disclaimed reliance on a formal “advice of counsel” defence at trial.^{[15, 16, 17]} The timing of a decision remains uncertain.

Executive Cooperation and Sentencing

The other members of the “inner circle” received significantly lighter sentences, reflecting (among other factors) extensive cooperation with prosecutors. Caroline Ellison, who testified that Bankman-Fried directed the misappropriation of funds, was sentenced to two years’ imprisonment; she reported to custody in November 2024, was transferred to community confinement in October 2025, and was released on January 21, 2026.^[18] Gary Wang and Nishad Singh avoided prison time and were sentenced to time served.^{[7, 19]}

In addition to criminal penalties, the SEC obtained final consent judgments against Ellison, Wang and Singh in December 2025, permanently enjoining them from violating antifraud provisions and imposing multi-year bans on serving as officers or directors of public companies.^{[20, 21]}

Remuneration and the Bankruptcy Estate: The Path to 119% Recovery

Despite the chaotic state of the company at the time of the filing, the bankruptcy estate has been remarkably successful in recovering assets. This success is attributed to the appreciation of crypto holdings, the liquidation of venture investments like the stake in Anthropic and the recovery of funds from political donations and real estate.^{[2, 22]}

The Payout Schedule and Creditor Controversy

The Chapter 11 reorganisation plan, approved in late 2024, became effective on January 3, 2025.^{[23, 24]} The plan specifies that 98% of creditors by number (the “Convenience Class” with claims under $50,000) will receive 100% of their allowed claim value plus a 9% annual interest rate, reaching a total of roughly 120% recovery.^{[24, 25, 26]}

While the nominal recovery figures are high, many creditors have expressed frustration. The payouts are based on the value of the crypto assets at the time of the bankruptcy filing in November 2022.^{[27, 30]} Because the broader crypto market has surged since that time, creditors are receiving only a fraction of what their tokens would be worth if they had not been liquidated. For example, a user who had 1 Bitcoin at the time of the crash, valued at ~19,000, while the market price of Bitcoin in early 2026 is exponentially higher.^{[22, 27]}

Regulatory Reconstruction: The Global Legislative Response

The systemic failure of FTX served as a major catalyst for intensified regulatory activity in digital finance. In the United States, the legislative response included enactment of the Guiding and Establishing National Innovation for U.S. Stablecoins Act (the GENIUS Act) on 18 July 2025, and House passage of the Digital Asset Market Clarity Act (the CLARITY Act), which (as of late 2025) remains pending in the Senate.^{[31, 32, 55, 56]}

The GENIUS Act: A New Standard for Stability

Signed into law on July 18, 2025, the GENIUS Act established the first federal regulatory framework for payment stablecoins.^{[32, 33, 34]} The act was designed specifically to prevent the type of commingling that occurred between FTX and Alameda.

1. 1:1 Reserve Mandate: Issuers must back all stablecoins 1:1 with high-quality liquid assets like U.S. dollars or short-term Treasuries.^{[32, 35]}

2. Segregation and Bankruptcy Priority: Reserve assets must be held in segregated accounts and are legally excluded from the issuer’s bankruptcy estate, giving holders priority over all other creditors.^{[33, 34, 36]}

3. Prohibitions on Interest: To maintain the status of stablecoins as payment tools rather than securities, issuers are prohibited from paying yield to holders.^{[33, 36]}

4. Enforcement and Fines: Wilful noncompliance can result in civil fines of up to $200,000 per day and criminal liability for executives.^[33]

The GENIUS Act’s principal requirements phase in through 2026–2027: the statute generally contemplates implementing regulations within one year of enactment and the effective date is tied to the earlier of January 2027 or a defined period after final rules are issued.^{[33, 35, 36]}

The SEC’s “Project Crypto” and Global Coordination

On 31 July 2025, SEC Chairman Paul Atkins announced “Project Crypto,” a Commission-wide initiative aimed at modernising securities regulation to support “on-chain” market infrastructure.^[57] In parallel, the European Union’s Markets in Crypto-Assets Regulation (MiCA) became fully applicable from 30 December 2024 (with stablecoin provisions applying from 30 June 2024), creating a harmonised licensing and “passporting” regime for crypto-asset service providers across the EU.^[58]

Technological Sovereignty: Proof of Reserves and ZK-Proofs

In the post-FTX landscape, the industry has adopted new technical standards to restore trust. The most significant of these is “Proof of Reserves” (PoR), a cryptographic process that allows users to verify that an exchange holds the assets it claims to hold.^{[37, 38]}

Technical Evolution of PoR

By 2025, PoR had evolved beyond simple snapshots. Some exchanges publish Merkle-tree-based attestations (Kraken), while others have begun to add zero-knowledge techniques (OKX, which states it uses zk-STARK) to prove liabilities without revealing individual balances.^{[37, 38]}

The use of zk-SNARKs (Zero-Knowledge Succinct Non-Interactive Argument of Knowledge) allows an exchange to prove it has a certain balance without revealing the underlying transaction details or user identities.^{[40, 41]} This addresses the “liabilities problem” mentioned by critics: it is not enough to show what you have; you must show what you owe. Modern PoR systems include margin accounts, futures holdings, and staked assets in their liability calculations.^[37]

Structural Lessons for the Global Financial Ecosystem

The fall of FTX provides crucial lessons for entrepreneurs, investors and regulators. It underscores the danger of “moving too fast” without the guardrails of traditional corporate governance.

The Failure of Due Diligence

Venture capital firms like Sequoia Capital, Paradigm and Softbank faced intense criticism for failing to identify glaring red flags at FTX. These included the lack of an independent board, the absence of a CFO and the use of “little-known” auditing firms like Prager Metis.^{[42, 43, 44]} The “bandwagon effect” and the fear of missing out (FOMO) led even sophisticated investors to ignore the fact that the company’s control environment was essentially non-existent.^{[45, 46]}

Auditor Liability and Independence

The SEC’s action against Prager Metis, resulting in a $1.95 million settlement, set a new precedent for auditor liability in the crypto space.^{[47, 48]} The firm was charged with negligence for failing to understand the FTX-Alameda relationship and for violating independence rules by including indemnification provisions in engagement letters for over 200 audits.^{[44, 48, 49]} The case highlights that industry-specific expertise is not optional when auditing complex digital asset platforms.^[49]

The Shift Toward Decentralisation

Perhaps the most enduring lesson is the growing preference for non-custodial solutions. The proportion of crypto spot trading conducted on decentralised exchanges (DEXs), measured by the DEX-to-CEX spot volume ratio, reached about 21% in late 2025.^[50] Traders, weary of centralised exchange failures, increasingly sought direct control over their assets through DeFi platforms and self-custody wallets.

The collapse of FTX served as a “leverage reset” that purged the market of speculative theatre and replaced it with systems that actually work.^{[51, 52]} While the human and financial cost was immense, the resultant regulatory and technological infrastructure has positioned the digital asset sector for a more stable and institutionally integrated future.

1. The collapse of FTX - KPMG agentic corporate services, https://assets.kpmg.com/content/dam/kpmg/cn/pdf/en/2022/11/the-collapse-of-ftx.pdf

2. The FTX Collapse: A Complete Guide - CoinLedger, https://coinledger.io/learn/the-ftx-collapse

3. Samuel Bankman-Fried Sentenced to 25 Years for His Orchestration of Multiple Fraudulent Schemes - Department of Justice, https://www.justice.gov/archives/opa/pr/samuel-bankman-fried-sentenced-25-years-his-orchestration-multiple-fraudulent-schemes

4. Sam Bankman-Fried’s FTX | MIT Sloan, https://mitsloan.mit.edu/sites/default/files/2024-06/Sam%20Bankman-Fried%27s%20FTX.pdf

5. IN THE UNITED STATES BANKRUPTCY COURT FOR THE DISTRICT OF DELAWARE In re: FTX TRADING LTD., et al.,1 Debtors. Chapter 11 Case No - AWS, https://pacer-documents.s3.amazonaws.com/33/188450/042020648197.pdf

6. Bankruptcy of FTX - Wikipedia, https://en.wikipedia.org/wiki/Bankruptcy_of_FTX

7. Bankman-Fried’s ex-deputy Wang avoids prison time over crypto fraud - Reuters, https://www.reuters.com/legal/bankman-frieds-ex-deputy-wang-avoids-prison-time-over-crypto-fraud-2024-11-20/

8. FTX’s downfall and Binance’s consolidation: The fragility of centralised digital finance - LSE Research Online, https://eprints.lse.ac.uk/119902/1/FTX_s_downfall.pdf

9. Sam Bankman-Fried - Wikipedia, https://en.wikipedia.org/wiki/Sam_Bankman-Fried

10. Crypto Comeuppance: A Deep Dive Into the Sentencing of FTX Founder Sam Bankman-Fried | Eye on Enforcement, https://www.eyeonenforcement.com/2024/04/crypto-comeuppance-a-deep-dive-into-the-sentencing-of-ftx-founder-sam-bankman-fried/

11. Culture, Controls, and Corporate Governance: Lessons from the FTX Fiasco | Walton College | University of Arkansas, https://walton.uark.edu/insights/posts/culture-controls-and-corporate-governance-lessons-from-the-ftx-fiasco.php

12. John J. Ray III - Wikipedia, https://en.wikipedia.org/wiki/John_J._Ray_III

13. FTX TRADING LTD., et al. Debtors’ Report on Failure of Internal Controls (Apr. 9, 2023) (Case 22-11068-JTD, Doc. 1242-1), https://www.fishmanhaygood.com/wp-content/uploads/2023/04/April-9-Debtor-Report-on-Failure-of-Internal-Controls-1.pdf

14. Sam Bankman-Fried sentenced to 25 Years on multi-billion-dollar fraud | WYPR, https://www.wypr.org/2024-03-28/sam-bankman-fried-sentenced-to-25-years-on-multi-billion-dollar-fraud

15. Sam Bankman-Fried’s Appeal: Key Dates, Arguments, and What’s Next - CCN.com, https://www.ccn.com/news/crypto/sbf-appeal-key-dates-arguments/

16. Sam Bankman-Fried asks Second Circuit for retrial of FTX fraud case, https://www.courthousenews.com/sam-bankman-fried-asks-second-circuit-for-retrial-of-ftx-fraud-case/

17. Appeals court skeptical of Sam Bankman-Fried’s bid to toss FTX fraud conviction - Reuters, https://www.reuters.com/legal/government/sam-bankman-frieds-lawyers-argue-new-fraud-trial-ftx-founder-2025-11-04/

18. FTX’s Caroline Ellison Set to Leave Prison Early - PYMNTS.com, https://www.pymnts.com/legal/2025/ftxs-caroline-ellison-set-to-leave-prison-early/

19. FTX co-founder Nishad Singh avoids prison after helping prosecutors - Reuters, https://www.reuters.com/legal/ftx-co-founder-nishad-singh-avoids-prison-after-helping-prosecutors-2024-10-30/

20. Caroline Ellison, Gary Wang, and Nishad Singh - SEC.gov, https://www.sec.gov/enforcement-litigation/litigation-releases/lr-26450

21. FTX execs settle with SEC and agree to officer-director bans - DL News, https://www.dlnews.com/articles/regulation/ftx-alameda-execs-agree-to-company-bars/

22. FTX to pay out additional $1.6 billion to creditors in third distribution | The Block, https://www.theblock.co/post/371486/ftx-to-pay-out-additional-1-6-billion-to-creditors-in-third-distribution

23. FTX Announces Effective Date and Record Date of January 3, 2025 for its Chapter 11 Plan of Reorganization - PR Newswire, https://www.prnewswire.com/news-releases/ftx-announces-effective-date-and-record-date-of-january-3-2025-for-its-chapter-11-plan-of-reorganization-302332816.html

24. FTX Creditors: Payment Timeline and What to Expect | Enes on Binance Square, https://www.binance.com/en/square/post/18419037908377

25. FTX Founder Sentenced, Bankruptcy Court Approves Reorganization Plan - Attorneys.Media, https://attorneys.media/ftx-sentencing-bankruptcy-plan/

26. FTX Repayment Progress Update! Large creditors will receive the remaining 27.5% principal, plus up to 80% interest compensation | Zombit on Binance Square, https://www.binance.com/en/square/post/26342493001897

27. FTX starts $5B payouts — Here’s what you need to know - TradingView, https://es.tradingview.com/news/cointelegraph%3A5eb33f286094b%3A0-ftx-starts-5b-payouts-here-s-what-you-need-to-know/

28. FTX set to repay USD 1.6 billion to creditors| The Paypers, https://thepaypers.com/crypto-web3-and-cbdc/news/ftx-set-to-repay-usd-16-billion-to-creditors-by-the-end-of-september-2025

29. FTX Recovery Trust to Distribute Approximately $1.6 Billion to Creditors in Third Distribution on September 30, 2025 - PR Newswire, https://www.prnewswire.com/news-releases/ftx-recovery-trust-to-distribute-approximately-1-6-billion-to-creditors-in-third-distribution-on-september-30--2025--302561856.html

30. FTX to start next creditor payout on Sep. 30 with $1.9 billion granted by court | The Block, https://www.theblock.co/post/364059/ftx-next-creditor-payout-sep-30

31. 2025 Crypto Regulatory Round-Up - Chainalysis, https://www.chainalysis.com/blog/2025-crypto-regulatory-round-up/

32. 2026 Crypto Regulation Outlook | The Block, https://www.theblock.co/post/383653/2026-crypto-regulation-outlook

33. Real GENIUS: Landmark U.S. Federal Payment Stablecoin Legislation | Winston & Strawn, https://www.winston.com/en/blogs-and-podcasts/non-fungible-insights-blockchain-decrypted/real-genius-landmark-us-federal-payment-stablecoin-legislation

34. Fact Sheet: President Donald J. Trump Signs GENIUS Act into Law - The White House, https://www.whitehouse.gov/fact-sheets/2025/07/fact-sheet-president-donald-j-trump-signs-genius-act-into-law/

35. GENIUS Act Is Officially Law — Now What? The Most Important Deadlines Stablecoin Businesses Should Know - Falcon Rappaport & Berkman, https://frblaw.com/genius-act-is-officially-law-now-what-the-most-important-deadlines-stablecoin-businesses-should-know/

36. The GENIUS Act: A New Era of Stablecoin Regulation - Gibson Dunn, https://www.gibsondunn.com/the-genius-act-a-new-era-of-stablecoin-regulation/

37. Kraken releases September 2025 Proof of Reserves, continuing our ..., https://blog.kraken.com/news/september-2025-proof-of-reserves

38. Proof of Reserves | Cryptocurrency Asset Verification | OKX (zk‑STARK description), https://www.okx.com/proof-of-reserves

39. A Comparative Analysis of zk-SNARKs and zk-STARKs: Theory and Practice - arXiv, https://arxiv.org/html/2512.10020v1

40. What Is ZK-SNARK? Cryptocurrency Use, Definition, and History - Investopedia, https://www.investopedia.com/terms/z/zksnark.asp

41. New regulations for cryptocurrency exchanges: a paradigm revolution in technical compliance, https://www.hashkey.com/en-US/blog/crypto-knowledge-hub/new-regulations-for-cryptocurrency-exchanges-a-paradigm-revolution-in-technical-compliance

42. Lessons to Learn from the Fall of FTX, https://www.nacdonline.org/all-governance/governance-resources/directorship-magazine/private-company-directorship-newsletter/lessons-to-learn-from-the-fall-of-ftx/

43. FTX – Lessons Learned from a Lack of Due Diligence - Azzad Asset ..., https://azzadasset.com/blog/ftx-lessons-learned-from-a-lack-of-due-diligence/

44. Prager Metis Settles SEC Charges for Faulty FTX Audits, Violations of Auditor Independence Rule - Thomson Reuters, https://tax.thomsonreuters.com/news/prager-metis-settles-sec-charges-for-faulty-ftx-audits-violations-of-auditor-independence-rule/

45. Cryptocurrency: Lessons From The Fall Of FTX | Oyster Consulting, https://www.oysterllc.com/what-we-think/cryptocurrency-lessons-from-the-fall-of-ftx/

46. 5 lessons to learn from the collapse of FTX. - Simon Business School, https://simon.rochester.edu/blog/deans-corner/5-lessons-learn-collapse-ftx

47. Prager Metis CPAs, LLC - SEC.gov, https://www.sec.gov/enforcement-litigation/litigation-releases/lr-26112

48. Audit Firm Prager Metis Settles SEC Charges for Negligence in FTX Audits and for Violating Auditor Independence Requirements - SEC.gov, https://www.sec.gov/newsroom/press-releases/2024-133

49. Auditor Negligence: Insights from the FTX & Prager Metis Case - Milosevic & Associates, https://www.mlflitigation.com/media/auditor-negligence-insights-from-the-ftx-and-prager-metis-case/

50. DEX to CEX Volume Ratios Reach New Highs in 2025 - CoinGecko Research (Nov. 28, 2025), https://www.coingecko.com/research/publications/dex-to-cex-volume-ratios-reach-new-highs-in-2025

51. How Crypto Rewired Itself for 2026 | DWF Labs, https://www.dwf-labs.com/research/from-4-billion-to-18-billion-in-rwas-and-50-stablecoin-growth-how-crypto-rewired-itself-for-2026

52. Crypto Price Forecast: Top Trends for Crypto in 2026 | INN - Investing News Network, https://investingnews.com/crypto-forecast/

53. Digital assets in 2026: From early days to regulated scale | Walkers - JDSupra, https://www.jdsupra.com/legalnews/digital-assets-in-2026-from-early-days-9657010/

54. $477 Million Moved out of FTX in Suspected Theft - Elliptic (Nov. 20, 2022), https://www.elliptic.co/blog/analysis/477-million-in-unauthorized-transfers-from-ftx

55. S.1582 — Guiding and Establishing National Innovation for U.S. Stablecoins Act (GENIUS Act) — Congress.gov, https://www.congress.gov/bill/119th-congress/senate-bill/1582

56. H.R.3633 — Digital Asset Market Clarity Act (CLARITY Act) — Congress.gov, https://www.congress.gov/bill/119th-congress/house-bill/3633

57. The SEC Launches “Project Crypto” — U.S. Securities and Exchange Commission (Aug. 5, 2025), https://www.sec.gov/about/sec-launches-project-crypto

58. European crypto-assets regulation (MiCA) — EUR-Lex summary (application dates), https://eur-lex.europa.eu/EN/legal-content/summary/european-crypto-assets-regulation-mica.html