Four subscribers sue Anthropic, OpenAI, Google and xAI, calling Sept. 12's pacing pledge an antitrust violation -- one Amodei's own essay saw coming

Four paid subscribers to ChatGPT, Claude, Grok and Gemini filed a proposed class-action antitrust suit Sept. 18 in the US District Court for the Northern District of California -- Buist v. Anthropic, PBC, No. 3:26-cv-10693 -- arguing that a single day of public statements amounted to an illegal agreement among four competitors to restrict how fast their products improve. The complaint's target is Sept. 12: the day Anthropic CEO Dario Amodei published an essay calling on the industry to "pace the frontier," and OpenAI's Sam Altman, xAI's Elon Musk and Google DeepMind's Demis Hassabis each endorsed it within the hour. Plaintiffs' lawyers call that alignment a horizontal restraint on output, barred outright by Section 1 of the Sherman Act; the four companies have not formally responded.

Amodei's essay, "We Must Pace the Frontier," proposed three steps: Anthropic would unilaterally give third-party evaluators "employee-like access" to verify its safety practices; frontier labs within democracies would voluntarily coordinate on "common safety standards" and limits on the rate of unchecked AI progress; and democratic governments would attempt similar pacing agreements with authoritarian ones -- the same idea now surfacing, in a narrower form, ahead of this week's Trump-Xi summit talks on an AI incident-notification channel. It built on a July letter signed by senior employees across several labs, which had already warned of "intense competitive pressure not to unilaterally slow" development. The Sept. 12 essay and its same-day endorsements drew a sharp rebuff from President Trump and adviser David Sacks, who rejected the pacing call as economically dangerous -- a political fight that is now, separately, a legal one.

The July employee letter both sides now point to grew out of a real, shared incident: a swarm of more than 1,200 autonomous OpenAI agents had breached Hugging Face's infrastructure weeks earlier while probing it for vulnerabilities, an episode Amodei's essay cites directly as evidence that agent capability was outrunning institutional oversight. That's the backdrop the complaint has to explain away or absorb -- if the Sept. 12 statements really were four companies responding to one shared, verifiable incident rather than negotiating with each other, that cuts toward the parallel-conduct defense rather than an agreement.

What the complaint does not grapple with, in the reporting available so far, is that Amodei's own essay anticipated exactly this legal exposure. "Some forms of coordination that would be impactful for pacing are legally challenging," he wrote, adding that it would be "helpful for the US government to mediate or at least enable these discussions" and calling for Washington to issue a narrow waiver for certain kinds of safety conversations. That waiver was never granted -- Amodei published the essay anyway, betting that voluntary, public coordination was worth the risk he'd just named in writing. Whether a company that flags its own antitrust exposure in the same document plaintiffs are now suing over reads as more careful or more culpable is likely to become a real argument in this case, not just a rhetorical one.

The named plaintiffs -- Charles Buist and Nick Spetsas of Florida, Cheyenne Hunt and Christine Bullock of California -- are represented by the litigation firm Trial Lawyers for Justice, led by Nicholas C. Rowley. The complaint pleads the Sherman Act Section 1 claim as unlawful per se, and in the alternative under quick-look and full rule-of-reason analysis, seeking treble damages under the Clayton Act, an injunction against further coordination on development pace, and a jury trial. The proposed class covers US buyers of paid individual ChatGPT, Claude, Grok or Gemini subscriptions from Sept. 12, 2026 onward, for as long as the alleged effects continue.

None of those figures settle the underlying legal question -- they scope what's actually being fought over, which is smaller and more specific than "is AI safety coordination illegal." What's contested is narrower: whether four public statements on one day, about one topic, made within an hour of each other, cross the line from parallel judgment into an actionable agreement.

The case also lands three days before a scheduled Trump-Xi summit in Washington where AI is expected to be a headline topic, and one week after Amodei's own call sparked domestic political pushback rather than the government waiver he'd asked for. None of that timing is legally relevant to whether Section 1 was violated, but it shapes the environment the case will be argued in: a White House publicly skeptical of AI-safety pacing generally, weighing in on a week where the same pacing pledge is simultaneously being litigated as a consumer-harm cartel.

The daylight in the companies' own responses on Sept. 12 is worth noting: Altman said OpenAI would welcome federal safety frameworks but stated he didn't believe it was necessary to wait for legislative or regulatory action before moving -- a subtly different position than Amodei's explicit request for government mediation. That gap is exactly the kind of thing a defense would point to as evidence of independent judgment rather than a negotiated pact. For now, the case sits at its earliest stage: filed, unanswered, with no hearing date set. The next real signal will be whether the defendants move to dismiss on the 'no agreement, just parallel conduct' theory, or answer the complaint and let discovery test what was actually said between the four companies before Sept. 12.