Sued for Going Slower: Inside the Antitrust Case That Turns AI Safety Pledges Into Sherman Act Evidence

On September 18, 2026, four consumers walked into the U.S. District Court for the Northern District of California in San Francisco and did something no one had done before: they sued the four biggest AI labs in America for agreeing to slow down.

The proposed class action targets Anthropic PBC, OpenAI OpCo LLC, SpaceXAI LLC (the entity through which Elon Musk controls the Grok business) and Google LLC. The claim is Section 1 of the Sherman Act — the same provision used to break up cartels — and the factual centerpiece is not a secret memo or a hotel-room handshake. It is a public essay, a stack of public endorsements, and a footnote that concedes the whole plan might need an antitrust waiver.

If the case survives a motion to dismiss, it will force courts to answer a question the AI industry has been dodging for a year: when four companies that control roughly 80% of the paid AI subscription market all publicly agree to improve their products more slowly, is that safety leadership — or price fixing by another name?

What the complaint actually alleges

The lawsuit was filed on behalf of a proposed nationwide class of U.S. residents who bought individual paid subscriptions to ChatGPT, Claude, Grok or Gemini from September 12, 2026 onward. The four named plaintiffs are Charles Buist and Nick Spetsas of Florida, and Cheyenne Hunt and Christine Bullock of California. Lead counsel is Nicholas C. Rowley of Trial Lawyers for Justice, joined by Andrew T. Tutt, R. Stanton Jones and Jakob Z. Norman. The case landed before U.S. Magistrate Judge Nathanael M. Cousins, and the plaintiffs have demanded a jury trial.

The core theory is quoted directly from the filing: “an agreement among rivals to reduce the quality of their products and the rate at which those products improve is an agreement to restrict output.” The complaint argues that this restraint does not become lawful just because the products happen to be new, or because the stated motivation is safety.

The theory of harm is unusually concrete for an antitrust case about a free-flowing software product. Subscribers, the plaintiffs say, kept paying the same monthly price while receiving products that improve more slowly than they would have under independent competition. Slower capability growth, framed as a lower-quality product sold at an unchanged price, becomes a quantifiable overcharge — the kind of injury antitrust law knows how to compensate, with damages that can be tripled under the Clayton Act.

The complaint is also careful about what it does not challenge. The plaintiffs explicitly state they take AI safety seriously, that guardrails should be set by the public through regulation and juries rather than by the defendants themselves, and that they are not contesting any defendant’s unilateral safety decisions or its advocacy to Congress. The claim is narrower and sharper: that what looks like four companies each independently choosing to slow down was, in fact, coordinated.

The timeline the plaintiffs built

The complaint assembles a chronology designed to move the four CEOs’ conduct from “parallel behavior” into “agreement” — the distinction Section 1 lives or dies on.

• July 2026: Representatives of Anthropic, OpenAI and Google form a working group to develop an industry standards body. On July 14, Google DeepMind co-founder Demis Hassabis publicly proposes a U.S.-led standards body modeled in part on FINRA, the financial industry’s self-regulatory organization.
• September 6: OpenAI chief scientist Jakub Pachocki publishes an essay, “An Alien Mind,” describing coordinated slowdown as one option available to frontier developers.
• September 10: WIRED reports that OpenAI had asked members of Congress whether an industry-wide coordinated slowdown would violate antitrust law — a signal, the plaintiffs argue, that the companies knew the legal territory was dangerous.
• September 11: In a Fortune interview, Sam Altman says he expects a common industry plan to emerge.
• September 12: Dario Amodei publishes “We Must Pace the Frontier,” writing: “We must slow the pace at which we improve the capabilities of AI models.” Within about an hour, Musk publicly endorses the proposal, Altman says he agrees and commits OpenAI to its first step, and Hassabis ties his endorsement to the standards body he proposed in July.
• September 14: Altman says AI progress would proceed more slowly and that OpenAI would not wait for an antitrust exemption or legislation before working with other labs.
• September 15: OpenAI global policy chief Chris Lehane confirms OpenAI had been working with Anthropic and Google DeepMind on the issue for several weeks.

The complaint also cites a separate July 2026 statement, “Pacing the Frontier,” carrying 1,386 signatories — including Amodei, Anthropic co-founders Jared Kaplan and Jack Clark, Pachocki, OpenAI chief research officer Mark Chen, and DeepMind’s Shane Legg — which asked governments to support an international pacing effort because of “intense competitive pressure not to unilaterally slow.”

The footnote that became an exhibit

The most striking detail in the filing is how much of the alleged “agreement” is public record. Amodei’s essay proposed three steps: embedded third-party evaluators with employee-like access (which Anthropic says it has adopted unilaterally); coordination among frontier developers in democratic countries on common safety standards and limits on the rate of unchecked progress; and, where possible, coordination extending to authoritarian governments.

A footnote in that same essay concedes the second step depends on government mediation or antitrust waivers — and urges companies to work together voluntarily in the meantime. That sentence, written as a pragmatic bridge, is now treated by the complaint as the blueprint the four defendants allegedly carried out.

That is the legal exposure: safety advocacy written in plain English becomes discovery material. Amodei himself flagged the antitrust risk in his own essay, and Altman mused publicly about a federal safety framework while noting companies could begin some of the work before an exemption existed. The plaintiffs’ argument is essentially: they knew, they did it anyway, and they said so in writing.

What happens next

None of the four defendants has yet filed substantive responses. The plaintiffs seek treble damages under the Clayton Act and an injunction barring horizontal agreements on AI development pace, training-compute limits, coordinated release delays, or exchanges of competitively sensitive information to police such a restraint. Notably, the proposed order would still allow each company to adopt its own safety measures, conduct research, and comply with government requirements — the plaintiffs are targeting coordination, not caution.

The case faces obvious hurdles. Section 1 requires proof of an agreement, and public endorsements of a safety philosophy are not contracts. Courts routinely dismiss cases where “conscious parallelism” — rivals independently responding to the same incentives — is the better explanation. And no lab has admitted to holding back a finished model.

But the case also arrives at a moment when the industry’s own statements have never been more specific. When CEOs testify that progress “would proceed more slowly,” when a policy chief confirms weeks of joint work, and when a 1,386-signature letter complains that competitive pressure prevents unilateral slowdown, the line between independent judgment and coordinated restraint gets blurry fast.

For an industry that has spent 2026 asking governments for antitrust exemptions to coordinate on safety, the lawsuit is a worst-case scenario realized: the pace-setting conversation itself, recharacterized as the conspiracy. Whatever Judge Cousins does with the motion to dismiss, the discovery phase alone — emails between rival policy teams, internal debates about release timing — could become the most detailed public record yet of how the frontier labs actually make decisions together.

The irony is thick enough to cut with a knife. The safety movement spent years arguing that AI development was too fast to govern. Now a federal courtroom will test whether slowing it down together was itself a form of market power abuse — and whether the price of safer AI was quietly paid by the subscribers who bought it.