Four AI giants sued over a 'pace the frontier' agreement
A class action filed in San Francisco on 18 September accuses Anthropic, OpenAI, SpaceXAI and Google of agreeing to slow the pace at which their rival AI products improve.
Four of the biggest names in artificial intelligence are being sued in San Francisco over what a new class action describes as an agreement between competitors to slow their own products down. The complaint, filed on 18 September in the Northern District of California, names Anthropic, OpenAI, SpaceXAI and Google as defendants.
The filing is Buist v. Anthropic, PBC, case number 3:26-cv-10693, assigned to Magistrate Judge Nathanael M. Cousins in the San Francisco Division. It runs to 29 pages and rests on a public exchange that began on 12 September, when Anthropic chief executive Dario Amodei published an essay titled “We Must Pace the Frontier”. Its central line, quoted in the complaint, is blunt: “We must slow the pace at which we improve the capabilities of AI models.”
What the plaintiffs say happened next
According to the complaint, the responses came the same day and in sequence. Elon Musk, who the filing says founded and controls the Grok business now named in court as SpaceXAI LLC, quote-posted the essay with “Dario is right”. OpenAI's Sam Altman wrote “I agree with Dario that we need to pace the frontier”. Google DeepMind co-founder Demis Hassabis called the essay “the right path forward” and tied it to the industry standards body he had proposed in July.
Two days later, the filing says, Altman spelled out what pacing meant: AI progress “should be slower than it otherwise could be”, and OpenAI would not wait for an antitrust exemption or legislation before getting started. On 15 September, the complaint says, OpenAI's global policy chief Chris Lehane confirmed that the company had been working on these questions with Anthropic and Google DeepMind for several weeks, and that it believed the firms could proceed without a waiver.
The complaint also points to a July statement signed by senior executives of Anthropic, OpenAI and Google, and to a working group of company representatives that it says has met regularly since July to build an industry standards body. It cites reporting that OpenAI asked members of Congress whether coordinating an industry-wide slowdown could breach the antitrust laws.
Why four subscribers can sue over it
The named plaintiffs – Charles Buist, Cheyenne Hunt, Christine Bullock and Nick Spetsas – are all paying subscribers to one or more of ChatGPT, Claude, Grok and Gemini. Their argument is that the value of those subscriptions is the steady stream of improvements, so an agreement that slows improvement leaves them paying for less than competition would have delivered. That, they say, is an overcharge of the kind antitrust law exists to prevent.
They are asking the court to certify a nationwide class of direct purchasers running from 12 September 2026, with separate subclasses for each defendant, to award damages trebled under the Clayton Act and to impose an injunction. The injunction they want would bar any agreement on the rate at which competing AI products are developed or released, on limits to training compute, on coordinated delays and on capability checkpoints.
The legal argument in one paragraph
The complaint's central claim is that the alleged arrangement is a naked horizontal restraint on output and product quality, unlawful per se under section 1 of the Sherman Act, with quick-look and rule-of-reason arguments pleaded in the alternative. It says the four companies account for at least 80% of paid consumer subscriptions to general-purpose frontier AI assistants, on information and belief, and that entry at the frontier is blocked by compute, talent and capital costs. It also argues that safety goals can be pursued unilaterally, so no pro-competitive justification applies.
The filing is careful about what it does not attack. It says it does not challenge unilateral decisions about safety, testing or the pace of any single company's own development, nor their right to lobby for regulation or an antitrust exemption. It challenges the agreement itself, which it says is being implemented outside any government process. None of the four companies had responded to the filing when this was written; the docket records a consent-or-declination deadline of 2 October and no hearing date.
Our opinion
The cleverest thing about this complaint is how little discovery it needs to look plausible. The agreement it alleges was proposed, accepted and confirmed in public, in writing, within about 72 hours, by chief executives who publish strategy as essays and then defend it on stage. That is a gift to a plaintiff's lawyer and an awkward position for four companies that spent September arguing that restraint is a virtue. Whether a court buys the per se framing is genuinely open – per se treatment is a high bar, and the rule-of-reason fallback drags everyone into market definition and expert economics for years. What the case settles already is narrower and more useful: when your business is watched this closely, a public plan for rivals to slow down together is a document that will eventually be read back to you under oath.