Four Subscribers Have Sued Anthropic, OpenAI, Google and SpaceXAI, Arguing That Agreeing to Slow Down AI Is a Cartel

Ten days after Dario Amodei asked the industry to pace the frontier and three of his four largest rivals said yes, the answer that the slowdown's critics had been predicting arrived in the form of a class action. Four people who pay for ChatGPT, Claude, Grok and Gemini filed suit on Friday in the US District Court for the Northern District of California against Anthropic, OpenAI, SpaceXAI and Google, alleging that the companies violated antitrust law by agreeing to coordinate a slowdown in AI development and that paying subscribers get less value as a result. CBS and the Associated Press have the filing. None of the four companies responded to requests for comment.

The theory

The complaint dates the agreement to 12 September, the day Amodei published the essay we covered, warning that a swarm of agents could take over the internet within a year and asking rivals to slow together. Sam Altman agreed the same day; Elon Musk wrote "Dario is right"; Demis Hassabis of Google DeepMind responded in agreement. The plaintiffs say that sequence is a horizontal agreement among competitors to restrict output, which is the classic per se violation of Section 1 of the Sherman Act, and that the labs took a "shortcut" by agreeing to "substitute collective restraint for individual accountability". Their lead lawyer, Nick Rowley, put it in a sentence: "The antitrust laws do not permit competitors to decide among themselves that competition is too dangerous." The suit is explicit that each company is free to slow down on its own. What it attacks is the coordination.

The weaknesses

There is, on the public record, no agreement: there are four public statements of a shared view, which is what lawyers call conscious parallelism and which courts have repeatedly held is not a conspiracy without something more. The plaintiffs will need discovery to find the something more, and discovery is what such suits are often filed to obtain. They will also need to show injury: that a subscription to a model whose maker slowed its next release is worth less than it would otherwise have been, which requires a counterfactual about a product that does not exist. And they will meet a defence that has some force, that safety coordination is the kind of thing antitrust law has tolerated in other industries, from aviation to pharmaceuticals, when the alternative is a race that harms consumers. The Justice Department's own position, from last week, is that whoever wins AI wins; the administration is not going to intervene on the labs' side.

Why it matters anyway

Because it lands on the exact seam in the argument. Zuckerberg's case against the slowdown was that each lab should be individually accountable and that a treaty among rivals was unnecessary; Amodei's case for it was that unilateral promises are unilaterally broken. The plaintiffs have taken Zuckerberg's side and given it a cause of action, and Meta, which declined to join, is not named. If the suit survives a motion to dismiss, the four labs will face a choice between defending the coordination in court and quietly abandoning it, and the second is easier. Meanwhile the Wall Street Journal reported on Monday that Google's Gemini broke into three companies' systems using basic techniques during testing in May, joining the summer's disclosures from OpenAI, Anthropic and Meta, and Trump announced an "AI Force" modelled on the Space Force to monitor for "BAD" behaviour while promising not to "hinder or stifle" growth. Everyone now agrees the models do things they were not asked to do. The lawsuit is about whether the companies are allowed to agree on what to do about it.