A Lawsuit Filed in San Francisco Says the AI Safety Consensus Is Illegal Under Antitrust Law

SAN FRANCISCO — A proposed class-action complaint against Anthropic and other AI companies begins with Anthropic chief executive Dario Amodei’s September 12 essay, “We Must Pace the Frontier.”

In it, Amodei acknowledged that coordination among frontier AI companies over development speed would require an antitrust waiver to be lawful. OpenAI chief executive Sam Altman subsequently said publicly that his company would not wait for such a waiver.

Six days later, four paying AI subscribers filed a class action in the U.S. District Court for the Northern District of California’s San Francisco Division. The complaint quotes both statements in its opening pages and presents them as evidence of potentially unlawful coordination.

The case is Buist v. Anthropic. The named plaintiffs are Charles Buist and Nick Spetsas from Florida, and Cheyenne Hunt and Christine Bullock from California, each a paying subscriber to one or more of the defendants’ products, according to CBS News. The defendants are Anthropic, OpenAI, SpaceXAI, and Google. The complaint does not dispute that any individual company has the right to make its own products safer by slowing capability advancement. The argument is that antitrust law does not allow competing companies to substitute what the complaint calls “collective restraint for individual accountability.”

The legal theory runs through Section 1 of the Sherman Antitrust Act, which prohibits contracts, combinations, and conspiracies in restraint of trade. Paying subscribers, the plaintiffs argue, are receiving slower-improving products at unchanged prices because their vendors reached an unlawful agreement. They seek treble damages, a permanent injunction, and a jury trial. The proposed class is nationwide.

Dario Amodei‘s September 12 essay is the evidentiary centerpiece. Within hours of publication, Sam Altman, Elon Musk of SpaceXAI, and Demis Hassabis of Google DeepMind each publicly endorsed the framework. The complaint treats the coordination of those endorsements, their simultaneous nature and alignment on a specific proposal by direct market competitors, as circumstantial evidence of an agreement. The July 2026 statement from high-ranking employees across multiple AI labs, expressing concern about “intense competitive pressure not to unilaterally slow,” is also in the complaint; the plaintiffs argue it demonstrates industry awareness that slowing down alone would cost market share, which means doing so together was not independent parallel behavior but collective action.

The specific problem for the defendants is one Amodei flagged himself. The essay explicitly stated that formal coordination among rival labs on AI development pace would require an antitrust waiver to be lawful, meaning the companies would need congressional authorization before agreeing to restrain competition. Sam Altman had told the Senate in July that AI might need to slow down, but his public position after September 12 was that OpenAI would not seek federal authorization before proceeding. The complaint treats that combination as an admission: Amodei identified the legal threshold; Altman said the industry would step over it anyway.

The defendants have substantial defenses. Parallel behavior among competitors is not illegal per se under antitrust law; companies frequently arrive at similar policies independently. The Noerr-Pennington doctrine shields joint efforts to influence government policy, and safety advocacy before international bodies can be framed as exactly that. A court has also never been asked to issue an injunction compelling companies to accelerate their own products. What that remedy would even look like in practice is a question the complaint leaves open. Whether the case survives a motion to dismiss will depend on whether the public endorsements, stripped of any formal meeting or written agreement, constitute a meeting of the minds sufficient to satisfy Section 1.

For Anthropic’s planned November IPO at a $2 trillion valuation, this is not an abstraction. An active federal antitrust conspiracy complaint against the company is a material disclosure item. The S-1 will have to address it, and underwriters will have to price the uncertainty.

The case lands in the same week that Sam Altman briefs the UN Security Council on AI governance, where he is arguing for exactly the kind of international coordination the lawsuit characterizes as illegal under domestic law. The gap between those two institutional settings is not a paradox but a structural problem: the United States has no mechanism for authorizing the international AI safety framework that multilateral bodies are building, short of the congressional antitrust waiver that Amodei mentioned and that neither Congress nor the Trump administration appears inclined to grant.

At the Trump-Xi summit on Thursday, Altman and other executives will sit alongside heads of state for discussions that include AI pacing. The domestic antitrust threat and the international diplomatic push are now occurring in the same week, with the same people at the center of both.

The Geneva process, BRICS, and the Security Council are each building architecture for exactly the kind of international pacing coordination that four American subscribers in federal court say is illegal under the Sherman Act, according to PBS News. Whether a multilateral safety framework can be built without a domestic legal vehicle to authorize it is a question no one has yet answered.

What the AI pacing consensus built in hours last week, this complaint may spend years unwinding in court.