Four Subscribers v. Four Labs: The Antitrust Case That Says Slowing AI Down Is a Cartel

On September 18, 2026, four people who pay monthly for ChatGPT, Claude, Grok and Gemini walked into a federal courthouse in San Francisco and accused the companies behind those products of running a cartel. Not a cartel to raise prices, and not a cartel to divide a market — a cartel, they allege, to slow down the pace of artificial intelligence itself.

The case is Buist et al. v. Anthropic PBC et al., No. 3:26-cv-10693, filed in the U.S. District Court for the Northern District of California. The named plaintiffs are paying subscribers to the four largest consumer AI services in the United States, and they bring the suit on behalf of a proposed nationwide class of paid subscribers to those same services. Their claim rests on Section 1 of the Sherman Antitrust Act, the 136-year-old statute that prohibits agreements between competitors that unreasonably restrain trade.

What the plaintiffs say happened

The complaint’s target is the “pace the frontier” movement that swept the industry in September 2026. On September 12, Anthropic CEO Dario Amodei published an essay titled “We Must Pace the Frontier,” arguing that AI labs should deliberately slow the rate of capability advancement in the interest of safety. He warned that swarms of rogue AI agents could take over critical internet infrastructure in as little as six months, and he proposed a three-part plan: independent safety verification, industry-wide coordination on pacing, and government mediation — including, he acknowledged, “a narrow waiver for certain kinds of safety conversations” to make the coordination legally safe.

Within days, OpenAI’s Sam Altman, SpaceXAI’s Elon Musk and Google DeepMind’s Demis Hassabis publicly endorsed the direction. OpenAI went further: policy chief Chris Lehane confirmed the company had already been working with Anthropic and Google DeepMind on safety coordination for several weeks, and WIRED reported that OpenAI had privately asked members of Congress whether an industry-wide slowdown would even be legal.

To the plaintiffs, that sequence is not public-spiritedness. It is the anatomy of an agreement. Paragraph two of the complaint puts it plainly: the alleged pact was “proposed in public, accepted in public, and confirmed in public.” The four CEOs’ essays, posts and interview comments are, in the plaintiffs’ telling, the offer and the acceptance.

A cartel case with no secret documents

What makes Buist unusual — and what makes it interesting far beyond the four defendants — is its evidence. There are no leaked emails in the filing, no confidential witnesses, no meeting minutes, and no informants. Every factual source is a named executive’s public statement or a press report about one.

That cuts both ways. It makes the allegations easy to verify: the “offer” is a published essay, the “acceptances” are public posts and interviews, and the confirmation of weeks of coordination came from OpenAI itself. It also means the alleged conduct is nothing anyone tried to hide — which is either evidence of innocence or, in the plaintiffs’ framing, evidence of how confident the companies had become.

The complained-of restraint is novel in form. Instead of agreeing on how many widgets to manufacture, the companies allegedly agreed on how quickly their AI systems would become more capable — through limits on training compute and training runs, restrictions on using AI to improve AI, and capability checkpoints policed by embedded evaluators. The claimed consumer injury is correspondingly unusual: subscribers continue paying the same subscription fees for products that will improve more slowly than they would have under unconstrained competition.

The law that got us here

The antitrust exposure was not a surprise to anyone involved. Before the lawsuit was filed, OpenAI had asked Congress for clarity on whether a coordinated slowdown would violate the Sherman Act — a question the company took seriously enough to seek legislative cover. Amodei’s essay itself conceded the point, acknowledging that some forms of coordination would be legally difficult and could require government involvement or an antitrust exemption.

The complaint’s answer to the waiver argument is flat: no such waiver exists, Congress has enacted no exemption, and no agency has compelled the conduct. Asking whether the conduct was lawful, the plaintiffs argue, demonstrates that the companies understood it as coordination — and proceeding anyway demonstrates intent. The suit also cites a July statement from employees across the labs acknowledging that each faced intense competitive pressure not to unilaterally slow. In antitrust terms, an admission that unilateral restraint is commercially irrational is precisely an explanation of why an agreement would be needed.

Crucially, the complaint does not allege that AI-safety concerns are invented. Its pretext argument is structural rather than personal: every stated safety objective, the plaintiffs say, can be pursued unilaterally, through independent evaluators, or through regulation. “The antitrust laws do not permit competitors to decide among themselves that competition is too dangerous,” the filing argues — a line that distills the entire case into one sentence. Lead attorney Nick Rowley put it more colorfully: “AI will quickly spin out of human control and could kill us all if we allow AI safety and protocol to be controlled by private self-serving agreements between the world’s most powerful ‘for profit’ technology companies.”

Where the case is vulnerable

The weaknesses are visible on the face of the complaint. It concedes that the full effect on released products has not yet manifested: no slowed release, canceled model, or altered training run is identified anywhere in twenty-nine pages. The claimed 80 percent share of U.S. paid subscriptions is pleaded on information and belief, as a floor rather than a measurement. And the four “acceptances” are not equivalent — Musk’s was four words on social media, while Google’s was an endorsement of direction with implementation left open.

Legal analysts note that under the Supreme Court’s Twombly standard, parallel conduct alone is insufficient; a plaintiff must supply additional facts making an agreement plausible rather than merely conceivable. Buist arguably clears that bar with its alleged working group operating since July, its closely timed public statements, and OpenAI’s own confirmation of weeks of talks — enough, perhaps, to survive a motion to dismiss and reach discovery. But plausibility is not proof, and an AI safety working group is not intrinsically an output cartel. The companies retain strong arguments that public advocacy and unilateral safety work are lawful, and that no defendant has responded yet.

Why it matters beyond the courtroom

The deeper significance of Buist is the question it forces: who decides how fast AI develops? The plaintiffs’ theory holds that private companies competing to build frontier intelligence cannot substitute their collective judgment about the desirable degree of competition for a decision that belongs to law or government. If safety-motivated coordination between competitors is legally hazardous, then every future attempt to build industry standards — embedded evaluators, shared red-teaming, common capability checkpoints — will need either a statute, a regulator, or a very careful lawyer.

The same question is now live in Europe, where Article 101 of the EU treaty prohibits agreements that restrict competition and contains no safety exemption a company can grant itself. And geopolitics pulls the other way: Beijing has already called the slowdown plan a “Cold War script,” meaning coordination is contested at both ends of the Pacific.

The immediate things to watch are procedural. The complaint alleges the cross-lab working group was still meeting in mid-September; continuing now would be read as continuing the conduct. Defendants’ responses and any motion to dismiss will arrive in the coming weeks. But win or lose, Buist has already changed the calculus: the four labs that spent September asking governments for permission to slow down now face a private lawsuit arguing they should have waited for the answer.