Consumers Sue OpenAI, Anthropic and Google for the Crime of Agreeing to Slow Down
A new antitrust lawsuit argues that AI labs publicly agreeing to be more careful is actually an illegal cartel, which means the industry has now been sued both for moving too fast and for saying it might not.
On September 19, four paying subscribers to ChatGPT, Claude, Grok, and Gemini filed a class-action lawsuit in the U.S. District Court for the Northern District of California against Anthropic, OpenAI, Google, and xAI — named in the complaint, somewhat gloriously, as “SpaceXAI.” Their claim is not that these companies are building AI too recklessly, too quickly, or with insufficient regard for the fate of humanity. Their claim is that four competing companies illegally agreed with each other to slow down, and that this agreement violates federal antitrust law by depriving paying customers of the products they would otherwise be racing to build.
This is, to be clear, a lawsuit alleging that a cartel formed for the purpose of restraining supply — except the supply being restrained is the pace at which machines that might end civilization get smarter. Lead attorney Nick Rowley’s own stated position, in the same breath as his antitrust theory, is that “AI will quickly spin out of human control and could kill us all if we allow AI safety… to be controlled by private self-serving agreements.” Rowley is arguing, simultaneously, that the danger is real and that the response to the danger is the crime.
The alleged conspiracy, per the complaint, has a start date: September 12, the day Anthropic CEO Dario Amodei published his “Pacing the Frontier” essay calling for the industry to voluntarily slow its own capability gains. Within hours, Sam Altman said he agreed. Elon Musk said Dario was right. Okay, Robots. covered this exact essay two weeks ago as a story about an industry agreeing with itself unusually fast. The lawsuit reads the same fourteen hours of public statements and sees something else entirely: three competitors, who spend the rest of their time suing each other over trade secrets and poaching each other’s researchers, allegedly forming an output-restricting agreement in full view of the press, using their own names, on a Saturday.
The complaint also points to a July 2026 statement, signed by senior employees at multiple labs, acknowledging that competitive pressure discourages any single company from unilaterally slowing down — the argument being that if the labs themselves admit competition is what keeps them racing, then a coordinated call to stop racing is functionally a price-fixing scheme with better PR. It is a clever legal theory. It is also a theory that requires treating “we should all be more careful” as economically indistinguishable from “we should all charge the same price for gasoline.”
The antitrust laws do not permit competitors to decide among themselves that competition is too dangerous, the plaintiffs argue — a sentence that could, with zero editing, serve as either a devastating legal citation or the tagline for a version of this industry that never gets built. There is something almost admirable about locating the one interpretation of the pacing essay in which everybody loses: if the labs go slow, safety advocates say it’s not slow enough and antitrust plaintiffs say it’s collusion; if the labs go fast, safety advocates say they’re gambling with the species. There is, per this lawsuit, no speed at which four AI companies can legally coexist while being simultaneously worried about their own product.
Whether the theory survives a motion to dismiss is a separate question from whether it’s being taken seriously. Sherman Act claims premised on public statements of shared concern — rather than secret meetings, back-channel pricing calls, or literal contracts — face a steep climb, and courts have generally required more than parallel conduct plus a plausible motive. Companies are allowed to agree publicly that an outcome is desirable without that agreement constituting a restraint of trade, provided nobody actually coordinated behavior in response. Whether four CEOs posting supportive replies to each other within a single afternoon crosses that line is now a federal judge’s problem, not a philosophy seminar’s.
None of the four defendants had commented on the suit as of this writing, which tracks: it is difficult to issue a statement responding to the accusation that you tried too hard to be responsible. Meanwhile, the underlying question — should the companies building potentially world-altering technology be allowed to talk to each other about doing it more carefully, and does the answer change depending on who’s asking — remains exactly as unresolved as it was on September 12. The only thing that’s changed is that it’s now also a docket number.
Sources: