The Agent That Wasn’t
AI-Washing and the Liability of the Decentralised Collective
A class action now before the Southern District of New York gathers two questions that the law of digital assets has not yet answered. The first is whether a project can be held to account for advertising machine autonomy it did not possess. The second is whether a decentralised autonomous organisation is a defendant that can be served, held to answer and made to pay. Pikabea v. Walters puts both in issue at once. The procedural skirmishing in its opening months shows why neither is straightforward.1
The complaint
The AI16Z token launched on the Solana blockchain in October 2024. It was marketed around an autonomous artificial intelligence agent said to make its own investment decisions and to manage an on-chain treasury. It also borrowed the cachet of Andreessen Horowitz, the venture firm whose a16z branding the project name evoked without any affiliation.2 By early January 2025 the token carried a market capitalisation of about $2.6 billion.3 It then collapsed. The token was later rebranded ELIZAOS.
Gorka Pikabea, a buyer resident in Spain, commenced proceedings in April 2026 on behalf of a proposed class of purchasers between 24 October 2024 and 20 April 2026. The complaint identifies at least 3,945 wallet addresses said to have sustained losses.4 Its central factual allegation is arresting in its simplicity. The autonomous agent, it says, was operated by hand. The machine agency that gave the token its narrative was, on the plaintiff’s account, a person.
The pleaded causes of action are reported as deceptive acts and false advertising under New York General Business Law, unfair competition and false advertising under California law, negligent misrepresentation and unjust enrichment.5 The named defendants are Eliza Labs, Inc., its founder Shaw Walters, Sebastian Quinn-Watson, the ai16z DAO itself and fifty unnamed persons. The relief sought includes damages, restitution and disgorgement. Every allegation remains unproven. A motion to dismiss was made in June 2026.6
AI-washing as a pleaded wrong
The choice of causes of action is the first point of interest. The plaintiff has not led with the federal securities laws. The wrong is framed as deception. The project said the product did something it did not do. That framing places the case in a lineage that predates the token boom and reaches well beyond it.
In March 2024 the Securities and Exchange Commission brought its first enforcement actions for what it called AI-washing, settling charges against two investment advisers, Delphia (USA) Inc. and Global Predictions Inc., for false and misleading statements about their use of artificial intelligence. Delphia had claimed to use an algorithm to analyse client data that it had not built. Global Predictions had described itself as the first regulated AI financial adviser. The firms paid $400,000 in total penalties without admitting or denying the findings.7 The regulatory point was that the misrepresentation of artificial intelligence is an old wrong wearing new clothes. A false statement about capability is actionable whether the capability claimed is human or machine.
Pikabea takes that logic into private litigation and into the crypto market. If the pleaded facts are made out, the deception is not in the token mechanics or the price action. It is in the representation of autonomy. That reframing carries consequences for proof and for reach. Consumer-protection and false-advertising claims turn on statements and their effect on the audience and do not carry the definitional battles over what is a security that have occupied the crypto docket for a decade. A claimant who can plead a clear misstatement of capability sidesteps the harder question of characterisation and gets to causation and loss.
The DAO as defendant
The second question is the one the court reached first and the more novel. The plaintiff named the ai16z DAO as a defendant in its own right, alongside the identifiable people behind it. That raises a problem the law of unincorporated collectives has been circling for several years. Who accepts service for an on-chain organisation with no registered office, no directors and no agent. At the initial pretrial conference the court questioned why the founder had not been served after a single attempt in San Francisco, noted that a co-defendant resides in Australia and heard the position, advanced for the defence, that the DAO could not be sued at all.8
The last submission is not fanciful, but it runs against the recent authorities. Two lines of cases now bear on whether a DAO is a defendant with legal capacity.
The first treats a DAO as an unincorporated association. In CFTC v. Ooki DAO the Commodity Futures Trading Commission obtained a default judgment against a DAO that had ignored the proceedings. Judge Orrick held that Ooki DAO was an unincorporated association subject to suit under the Commodity Exchange Act, entered judgment for $643,542, imposed trading and registration bans and ordered the associated website content removed.9 The ruling’s importance lay in a single holding, that the collective was a person the statute could reach.
The second line is more consequential for members and runs through the general partnership. In Sarcuni v. bZx DAO the Southern District of California declined to dismiss a negligence claim brought by users who had lost about $1.7 million in a protocol hack, finding that the plaintiffs had sufficiently alleged that bZx DAO was a general partnership under the California Corporations Code and that its members were the partners.10 A general partnership carries joint and several liability. On that theory the liabilities of the collective become the personal exposure of those who compose it.
The general partnership problem
The general partnership theory was sharpened a year later. In Samuels v. Lido DAO, a claim that the LDO token was an unregistered security, Judge Chhabria held that Lido DAO could be a general partnership and, more pointedly, that certain institutional participants were plausibly its general partners. The court distinguished passive holding from active governance. Paradigm Operations, Andreessen Horowitz and Dragonfly, which held large blocks of governance tokens and had held themselves out as active in the organisation, were plausibly partners. Another investor, Robot Ventures, was dismissed for want of sufficient allegations of active participation.11 The general counsel of a16z crypto called the decision a serious blow to decentralised governance, on the view that any participation, down to a forum post, might fix a member with the liabilities of the whole.12
That distinction is the legal core of the coming fight and disciplines the loose language that surrounds these cases. Liability does not attach to token holders as such. It attaches to members or active participants, to those who exercise or hold out that they exercise control. The question in each case is not who owned a token but who ran the enterprise. For ai16z, where the allegation is precisely that a human hand directed what was sold as an autonomous system, the general partnership analysis and the AI-washing analysis converge on the same small group of people.
What it means
Three consequences follow for anyone building or funding in this space.
For AI-token projects the exposure is not confined to the collapse of the token. It sits in the marketing. A project that advertises autonomy it does not have invites a deception claim that may be easier to plead than a securities case because it does not first require the claimant to win the token-classification fight. The Commission has already shown that a regulator will treat the misdescription of artificial intelligence as a discrete wrong. The gap between what a system is said to do and what it does is now a legal surface.
For DAOs the case tests whether the structure absorbs a claim or merely makes the defendants harder to find. The recent authorities suggest the latter. A court that regards a DAO as an unincorporated association may let it be sued. A court that regards it as a general partnership may look through it to the participants who directed it. Decentralisation complicates service. It does not, on the current cases, dissolve liability.
For founders and insiders the practical lesson is that the collective form is not obviously a shield. The claimant strategy these cases invite is straightforward. Name the reachable humans, plead the general partnership to widen the net and let the difficulty of serving a decentralised entity fall on the defence rather than the claim.
The open question
What remains unresolved is whether the merits will be reached at all. The motion to dismiss will test whether AI-washing dressed as consumer protection states a claim. The DAO-capacity issue may arise separately through service, representation and any later challenge to the DAO’s amenability to suit. But the case may turn on service and settlement before it turns on doctrine. If the DAO cannot be served or represented, the claim falls back onto the identifiable people the court can reach. The fork the court has not resolved is whether naming the DAO widens liability to its members or active participants or simply exposes how little of a decentralised collective the law can, in practice, touch.
Code on Trial is a weekly briefing on AI, crypto and digital disputes.
Notes
1 Pikabea v. Walters et al, No. 1:26-cv-03238 (S.D.N.Y.), before Judge Rakoff. Docket via Justia and PacerMonitor. All allegations unproven.
2 Cointelegraph (22 May 2026); secondary crypto reporting on the launch and marketing of the AI16Z token.
3 Cointelegraph, 22 May 2026 (peak market capitalisation of about $2.6 billion, January 2025).
4 Complaint as summarised in secondary reporting (Claim Depot; KuCoin). The class period and the figure of at least 3,945 wallet addresses should be confirmed against the complaint.
5 Pleaded counts per secondary summaries of the complaint: New York General Business Law ss 349 and 350; California Unfair Competition Law and False Advertising Law; negligent misrepresentation; unjust enrichment.
6 Docket: motion to dismiss made 18 June 2026; opposition due 3 July; replies due 13 July 2026; telephonic pretrial conference listed 31 July 2026.
7 SEC, Press Release 2024-36 (18 March 2024); settled orders In re Delphia (USA) Inc. and In re Global Predictions Inc.; combined penalties of $400,000.
8 Inner City Press, report of the 19 May 2026 conference before Judge Rakoff; Cointelegraph, 22 May 2026.
9 CFTC v. Ooki DAO, No. 3:22-cv-05416-WHO (N.D. Cal.), default judgment 8 June 2023 (Orrick, J.); judgment for $643,542 with trading and registration bans and removal of website content.
10 Sarcuni v. bZx DAO, No. 3:22-cv-00618, 664 F. Supp. 3d 1100 (S.D. Cal. 2023), order of 27 March 2023 (partial denial of motions to dismiss).
11 Samuels v. Lido DAO, No. 3:23-cv-06492 (N.D. Cal.), order of 18 November 2024 (Chhabria, J.); Securities Act 1933 s 12(a)(1) claim; general partnership finding as to certain participants.
12 Miles Jennings, General Counsel, a16z crypto, public commentary on the Lido decision (reported by Decrypt, November 2024).