On 24 August 2026, a federal jury in San Francisco convicted Japheth Dillman, the founder of cryptocurrency investment fund Block Bits, of wire fraud and conspiracy to commit wire fraud. The prosecution was framed as a crypto case. Its more important feature was simpler. Investors were told that a proprietary automated trading system existed, worked and was producing returns. The government proved that it did not.1
The verdict is an early and unusually clean example of what now appears across artificial intelligence markets as “AI-washing”: the sale of an investment proposition by attributing capability, activity or performance to technology that is absent, unfinished or materially different from the system described. Technology changes. The legal problem does not. A statement about an algorithm is a statement of fact when it asserts that the algorithm has been built, deployed or used to generate results.
Block Bits therefore reaches further than the modest sum raised. The case sets out the evidential anatomy of technology fraud. The enquiry separates existence, capability, deployment and performance attribution. A promoter who compresses those four questions into the word “proprietary” does not make them less susceptible to proof.
The representation was the product
Dillman and his co-founder, David Mata, began soliciting investment in 2017. According to the Securities and Exchange Commission’s civil complaint, the fund was marketed as “An Automated Cryptocurrency Fund”. Its proposed advantage was an in-house trading bot capable of arbitrage across approximately 100 cryptocurrencies and 30 exchanges. The SEC alleges that Dillman told prospective investors that the system had been built and, on 25 September 2017, that “We have finished the Arbitrage AutoTrader”.2
Those details come from a civil pleading and remain allegations in that proceeding. The criminal verdict establishes the central proposition on a different and higher standard of proof. The Department of Justice said the trial evidence showed that Dillman raised nearly $1 million from more than 20 investors while representing that the Autotrader was working and complete. He knew it was not working and that investor money could not be used in the manner promised.3
A conviction is not proof of every allegation made by another agency in a parallel case. It is proof beyond reasonable doubt of the offences submitted to the jury. The public verdict announcement records convictions for wire fraud and conspiracy to commit wire fraud after a ten-day trial before U.S. District Judge Richard Seeborg. Law360 reported that the jury returned guilty verdicts on four wire-fraud counts and one conspiracy count after approximately six hours of deliberation over two days.4
Mata testified for the prosecution under a plea agreement. He had pleaded guilty in June 2022 to one count of wire fraud arising from the Block Bits scheme.5 Co-operator evidence often becomes the bridge between promotional language and internal knowledge. In a technology case, that bridge is particularly important. The external claim may be expressed through a pitch deck, an email or a performance report. The system’s actual condition appears in the source code, the testing records, the exchange accounts, the development messages and the evidence of the person expected to make it work.
Four falsifiable claims
“We use an algorithm” sounds general. In an investment solicitation it can contain at least four distinct representations.
The first concerns existence. Was there identifiable software rather than an aspiration, specification or unfinished development project? The second concerns capability and asks whether the software had been tested sufficiently to support the functions attributed to it. Deployment comes third. Was the system connected to live accounts and used to execute the relevant trades? The fourth question is attribution and asks whether reported returns were generated by that system rather than by manual trading, market appreciation, selective accounting or an unrelated investment.
The SEC’s allegations illustrate the separation. On its case no functional bot was ever tested or deployed and Mata conducted the fund’s trading manually. Investor reports nevertheless continued to describe automated arbitrage and remarkable returns. If proved, the problem would not be that the code performed poorly. It would be that the claimed technological process and the actual investment process were different things.6
This is why the familiar defence that startups sell a future is incomplete. There is a legal and factual difference between a development plan and a statement that development is complete. “We intend to build” describes a proposal. “We have finished” describes an existing fact. The first may fail commercially. The second may be fraudulent when it is knowingly false, material and used to obtain money. Nor is a statement of intention immune. It may be fraudulent where the speaker did not hold the stated intention, or knowingly presented the proposed outcome as attainable when there was no basis for doing so. Someone who says he intends to build, knowing there is neither funding nor a developer, stands where the promoter who says he has finished stands.
The distinction also answers a recurring objection in cases involving opaque systems. A jury need not evaluate the elegance of the code or determine whether one technical architecture was preferable to another. Jurors can compare what investors were told with what existed, what was deployed and what happened to their money. Complexity may enlarge the evidential record. It does not displace the comparison.
Legacy fraud law, modern evidence
Federal wire fraud does not require a special crypto or algorithmic offence. The prosecution must prove a scheme to defraud resting on a material misrepresentation, use of interstate wires in furtherance of that scheme and specific intent to defraud. Intent may be established through circumstantial evidence.7 Conspiracy under 18 U.S.C. § 1349 addresses the agreement to commit the underlying offence and, unlike a general conspiracy, requires no overt act and carries the same maximum penalty as the substantive offence.8
On the government’s account, the relevant falsehoods were material because automation was not a peripheral feature of the offering. It was the stated investment method. Block Bits was not merely buying digital assets while using ordinary software in the background. Its claimed advantage was the capacity to identify and execute arbitrage opportunities across fragmented markets without relying on manual intervention.
The use of old law is a strength rather than an embarrassment. Technology-specific legislation is necessary when the conduct creates a new interest or harm. It is unnecessary when novel terminology is used to obtain familiar property by familiar deceit. No statutory definition of an autotrader is needed before a jury may decide whether a representation that it was complete was false.
The evidence, however, is technologically specific. A serious diligence or enforcement exercise should preserve the development repository, release history, test outputs, application logs, API connections, exchange credentials, trade timestamps and model or strategy documentation. Promotional claims should then be mapped against those records. The decisive document may be neither the source code nor the pitch deck, but the version history showing that the represented function did not exist on the date money was solicited.
Performance attribution requires the same discipline. If a manager says a system generated a return, the assertion should be capable of reconciliation to live-account logs, executed orders, fees, cash movements and valuation records. A back-test is not live trading. A gross return is not a net investor return. Manual selection of a favourable transaction produces neither an automated output nor a representative one. A model that recommends a trade may sit some distance from the system that executes it.
From crypto-washing to AI-washing
The Block Bits conduct predates the present investment fashion for artificial intelligence, but the representational pattern is current. In March 2024, the SEC settled enforcement proceedings against two investment advisers, Delphia (USA) Inc. and Global Predictions Inc., over false or misleading statements about their use of AI. The Commission described the conduct as AI-washing. The firms agreed to pay a combined $400,000 in civil penalties without admitting or denying the findings.9
The analogy should not be overstated. Those were civil and administrative settlements. Dillman has been convicted by a jury of federal crimes. The underlying compliance lesson is nevertheless common. A technology claim must be supported at the level at which it is made. If marketing says that AI drives portfolio construction, it is insufficient that the firm owns analytical software, has experimented with a model or intends to incorporate machine learning later.
Prosecutors have since brought a criminal case of the same kind on the artificial intelligence side. In April 2025 the Securities and Exchange Commission sued Albert Saniger, founder and former chief executive of Nate, Inc., over a shopping application marketed as completing purchases through proprietary artificial intelligence. The Commission alleges that he raised more than $42 million from investors while the application relied in large part on contract employees who entered orders by hand. The Department of Justice charged him in parallel in the Southern District of New York. Block Bits and Nate share a structure. Each rests on a claimed automated system, on human execution behind it and on investor money raised on the representation.10
The same four questions apply. Does the model exist? Does it possess the described capability? Was it used in the relevant decision process? Did it generate the result attributed to it? For generative AI, the enquiry extends to which model and version operated at the relevant time, what data and tools were available to it, whether a human decision was presented as an autonomous output and whether favourable examples were selected from a wider set of failed or discarded ones.
These are not semantic objections. They determine whether the buyer received the process for which capital was sought. They also determine scienter. Internal model cards, validation reports, deployment approvals and incident logs may show uncertainty or limitation. That does not itself establish fraud. The sharper evidence is a contradiction between those internal records and a confident external assertion that the limitation does not exist.
The money trail still decides cases
Technology representations do not replace conventional financial analysis. They direct it. The Department of Justice said evidence showed that investor money was used to pay the founders and to pursue risky and speculative crypto ventures, while investors received misleading reports that concealed losses.11
The SEC’s complaint supplies further allegations. On the Commission’s case approximately $960,000 was raised from approximately 22 retail investors and funds Dillman described as sitting in risk-free cold storage were in fact placed in high-risk loans and token investments, among them the AML Bitcoin offering the Commission has separately alleged to be fraudulent. It alleges that approximately $300,000 was transferred to the general partner, including $67,500 that reached Dillman personally. The Commission characterises much of the compensation as unauthorised under the offering terms.12 Those propositions remain for determination in the civil case unless resolved by agreement or otherwise disposed of.
The civil proceedings have been stayed while the criminal case proceeds. An order of 8 July 2022 required the parties to request a status hearing within thirty days of the resolution of the criminal case.13 The verdict materially changes the posture of the SEC case and may narrow the issues remaining for determination, but it does not automatically decide the civil proceedings. Sentencing is listed for 8 December 2026. Post-trial motions and any appeal remain possible.14
The compliance consequence
Boards, investment committees and counsel should treat material technology claims as controlled statements. Each claim should have an owner, a dated evidential basis and a defined scope. “Uses AI”, “automated”, “real time” and “proprietary” are not harmless adjectives. They are compressed factual propositions.
The control should extend beyond marketing approval. Development and investment records must use the same vocabulary. If engineers describe a system as a prototype while fundraising materials call it operational, the discrepancy requires resolution before publication. If performance reports attribute returns to an algorithm, the attribution requires account-level support. If human intervention is substantial, the disclosure should say so.
Block Bits does not extend fraud law into an unregulated technological frontier. It shows that no frontier needed to be crossed. An unfinished trading bot was presented as a completed investment engine. Investors supplied money. The wires carried the representations. The internal and financial records supplied the contradiction.
A promoter can obscure a product behind the vocabulary of innovation for a time. He cannot change the legal character of a false statement about what has been built, what has been used and what has produced the return.
Notes
1. U.S. Attorney’s Office, Northern District of California, “Founder of Cryptocurrency Trading Fund Convicted of Defrauding Investors” (24 August 2026), https://www.justice.gov/usao-ndca/pr/founder-cryptocurrency-trading-fund-convicted-defrauding-investors.
2. Complaint, SEC v Block Bits Capital, LLC et al., No. 3:22-cv-02563 (N.D. Cal., filed 28 April 2022), paras 1–5, 14–15, 18, https://www.sec.gov/files/litigation/complaints/2022/comp25376-blockbits.pdf.
3. U.S. Attorney’s Office, note 1.
4. Bonnie Eslinger, “Founder Of Crypto Startup Block Bits Convicted Of Fraud”, Law360 (24 August 2026), https://www.law360.com/fintech/articles/2516827; United States v Japheth Dillman, No. 3:23-cr-00140 (N.D. Cal.).
5. Eslinger, note 4 (Mata pleaded guilty in June 2022 and testified for the prosecution). For the charge, see U.S. Attorney’s Office, Northern District of California, “San Francisco Man Charged In Alleged Cryptocurrency Investor Fraud Scheme” (27 April 2022), https://www.justice.gov/usao-ndca/pr/san-francisco-man-charged-alleged-cryptocurrency-investor-fraud-scheme; SEC, “SEC Charges Promoters of ‘Automated Cryptocurrency Fund’ with Fraud and Registration Violations”, Litigation Release No. 25376 (28 April 2022), https://www.sec.gov/enforcement-litigation/litigation-releases/lr-25376. Mata was charged by separate information on the same day.
6. SEC Complaint, note 2, paras 3–5, 17, 20–21.
7. 18 U.S.C. § 1343; United States v French, 748 F.3d 922, 935 (9th Cir. 2014) (scheme, wires and specific intent; intent may be circumstantial), https://cdn.ca9.uscourts.gov/datastore/opinions/2014/04/07/12-10185.pdf; Neder v United States, 527 U.S. 1, 25 (1999) (materiality is an element of the federal fraud statutes).
8. 18 U.S.C. § 1349, which provides the same penalties as the substantive offence. A general conspiracy under 18 U.S.C. § 371 requires an overt act. Where a conspiracy statute is silent as to an overt act, none is required. See Whitfield v United States, 543 U.S. 209, 213–214 (2005), applying United States v Shabani, 513 U.S. 10 (1994).
9. SEC, “SEC Charges Two Investment Advisers with Making False and Misleading Statements About Their Use of Artificial Intelligence” (18 March 2024), Release No. 2024-36, https://www.sec.gov/newsroom/press-releases/2024-36.
10. SEC, “SEC Charges Founder and Former CEO of Artificial Intelligence Startup with Misleading Investors”, Litigation Release No. 26282 (11 April 2025), https://www.sec.gov/enforcement-litigation/litigation-releases/lr-26282; SEC v Saniger Mantinan, No. 1:25-cv-02937 (S.D.N.Y., commenced 9 April 2025). For the criminal charges see U.S. Attorney’s Office, Southern District of New York, “Tech CEO Charged In Artificial Intelligence Investment Fraud Scheme” (9 April 2025), https://www.justice.gov/usao-sdny/pr/tech-ceo-charged-artificial-intelligence-investment-fraud-scheme (one count of securities fraud and one count of wire fraud).
11. U.S. Attorney’s Office, note 1.
12. SEC Complaint, note 2, paras 4–5, 27.
13. Stipulation and Order Re: Stay of Proceedings, SEC v Block Bits Capital, LLC et al., No. 3:22-cv-02563-RS (N.D. Cal., 18 August 2022), https://www.govinfo.gov/content/pkg/USCOURTS-cand-3_22-cv-02563/pdf/USCOURTS-cand-3_22-cv-02563-1.pdf.
14. U.S. Attorney’s Office, note 1.
Image: photograph by Anne Nygård, Unsplash.