On Friday, September 18, four AI subscribers filed a proposed class-action lawsuit in the U.S. District Court for the Northern District of California accusing Anthropic, OpenAI, SpaceXAI, and Google of something that would have sounded absurd five years ago: agreeing not to compete too hard. The complaint argues the four frontier AI labs violated Section 1 of the Sherman Act by coordinating a slowdown in the pace of AI capability improvements — harming paying subscribers of ChatGPT, Claude, Grok, and Gemini, who are getting less innovation for their money.

The case is remarkable for what it does not allege. There is no secret pricing cartel, no backroom deal, no leaked memo. The alleged agreement happened in full public view, across a single week of essays, interviews, and social-media replies.

What the complaint actually claims#

The four named plaintiffs — Charles Buist and Nick Spetsas of Florida, and Cheyenne Hunt and Christine Bullock of California — each pay for AI subscriptions. Buist, Hunt, and Spetsas subscribe to all four services (ChatGPT, Claude, Grok, and Gemini); Bullock subscribes to Claude. They sue on behalf of a proposed nationwide class of all U.S. consumers who bought a paid individual subscription to any of the four services from September 12, 2026 onward.

The defendants are Anthropic PBC, OpenAI OpCo LLC, SpaceXAI LLC, and Google LLC — described in the filing as the four companies at the frontier of AI.

The legal theory is stated flatly. The complaint calls the arrangement an agreement among chief rivals that their progress "should be slower than competition would otherwise produce," harming consumers. "The antitrust laws do not permit competitors to decide among themselves that competition is too dangerous," the plaintiffs write. And: "Plaintiffs challenge only what the antitrust laws forbid: an agreement among competitors about how fast their competing products will improve. Congress has granted no exemption for that agreement."

The pleading frames the restraint as unlawful per se and, in the alternative, under quick-look and rule-of-reason analysis.

The week that built the case#

The complaint's timeline is unusually tidy — nearly every link in it is a public document:

  • 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 for frontier AI, modeled in part on the Financial Industry Regulatory Authority. A statement titled Pacing the Frontier, supported by the nonprofits Guidelight AI Standards and Encode AI, gathers 1,386 signatories — including Amodei, Anthropic co-founders Jared Kaplan and Jack Clark, OpenAI's Jakub Pachocki and Mark Chen, and Google DeepMind co-founder Shane Legg. The signatories describe an "intense competitive pressure not to unilaterally slow" capability development and ask the U.S. government to support an international effort to deliberately pace frontier AI.
  • September 6. OpenAI publishes an essay by chief scientist Jakub Pachocki, "An Alien Mind," describing coordination among frontier developers to slow future development as one of the principal options. Days later, WIRED reports that OpenAI has asked members of Congress whether coordinating an industry-wide slowdown could violate antitrust law.
  • September 11. In a Fortune interview, OpenAI CEO Sam Altman says he expects a common industry plan to happen, while declining to detail private discussions.
  • September 12. Anthropic CEO Dario Amodei publishes the essay We Must Pace the Frontier: "We must slow the pace at which we improve the capabilities of AI models." Within about an hour, the complaint alleges, Elon Musk publicly endorses the proposal, Altman writes that he agrees and commits OpenAI to the plan's first step, and Hassabis endorses the essay's direction, tying it to the standards body he proposed in July.
  • September 14. Altman states that AI progress will proceed more slowly than it otherwise could, and that OpenAI will not wait for an antitrust exemption or legislation before beginning work with colleagues across the industry.
  • September 15. OpenAI global policy chief Chris Lehane confirms OpenAI has been working with Anthropic and Google DeepMind on these issues for weeks.
  • September 18. The class action is filed. Lead attorney Nicholas C. Rowley of the firm Trial Lawyers for Justice argues the alleged pact would let "AI safety and protocol to be controlled by private self-serving agreements between the world's most powerful 'for profit' technology companies."

As of Saturday, none of the four companies had responded to requests for comment, according to multiple outlets.

Why antitrust law may not care that this is about safety#

Here is the uncomfortable core of the case: antitrust law cares whether competitors coordinated, not why. An agreement to restrain output does not become lawful because the restraint is in a good cause. A genuine joint safety standard set by regulators would sit in a different category — but a self-organized agreement among for-profit competitors to slow the rate at which their products improve looks, on the plaintiffs' telling, like a classic output-restricting cartel.

The plaintiffs say they take AI safety seriously but believe guardrails should be set by the public through regulation and juries, not by the defendants. They expressly do not challenge any defendant's unilateral safety decisions, the pace of its own development, or the companies' advocacy to Congress or the White House. That carve-out is strategic: it keeps the case about agreement among competitors — what Section 1 targets — and concedes everything else.

The consumer-harm theory is equally straightforward. Subscriptions to ChatGPT, Claude, Grok, and Gemini are marketed and priced on access to each company's most capable models and continuing improvements — OpenAI's ChatGPT Plus tier is priced at approximately $20 per month — so an agreement to slow improvement lowers the quality subscribers receive for the price they pay. The filing calls this an overcharge of the kind antitrust law was enacted to prevent, and alleges that the four defendants collectively account for at least 80 percent of paid consumer subscriptions in that U.S. market.

The defendants' problem: the evidence is their own marketing#

What makes this complaint sharper than a typical parallel-conduct case is that the plaintiffs are not reconstructing a hidden agreement from indirect evidence. They are pointing at an essay and three public replies and arguing the pattern itself is the violation. Amodei's own essay anticipated antitrust scrutiny — a footnote concedes that the proposed industry coordination depends on government mediation or antitrust waivers.

That self-awareness cuts both ways. It shows the defendants knew exactly which legal line they were near — which plaintiffs will read as a knowing agreement — and it shows the labs believed a government-mediated process was the legitimate route. The question for a judge: do four public statements on the same day amount to an "agreement" under antitrust law, or four companies independently reaching similar conclusions and saying so out loud? Courts have historically been cautious about treating parallel public statements, without more, as proof of an illegal agreement. Whether "please can we all agree to go slower" crosses the line into an actionable restraint of trade is genuinely open — and this case is the vehicle testing it.

What the plaintiffs want, and what happens next#

The relief sought is ambitious: treble damages under the Clayton Act and an injunction barring each defendant from any horizontal agreement concerning the rate at which competing AI products are developed, improved, trained, or released — including limits on training compute, coordinated release delays, capability checkpoints, or exchanges of competitively sensitive information to police such a restraint. The plaintiffs also demand a trial by jury.

The next milestone is a motion to dismiss, where the defendants will likely argue the complaint alleges parallel public statements, not an actual agreement to restrain output. If the case survives that, discovery would probe whether private communications preceded or followed the public exchange — a paper trail beyond the essay and the replies would matter far more than the essay itself. A parallel government track is also possible: the FTC and DOJ have both been active on AI competition this year, and a private suit built around a clean public timeline is exactly the kind of case that can draw a regulator's attention even while the private suit moves slowly.

Things to watch: the first company statement, the motion to dismiss, any FTC or DOJ interest, and copycat suits from other subscribers or state attorneys general now that the theory is public.

Takeaway#

This lawsuit is the first real stress test of a collision the AI industry has been walking toward for years: frontier labs want to coordinate on safety, and antitrust law forbids competitors from coordinating on output. The labs' own caution — the antitrust questions asked of Congress, Amodei's waiver footnote — shows they saw the collision coming. Now a court will decide whether public agreement to slow down, however well-motivated, is an illegal agreement to restrain trade. Whatever the answer, every future conversation between competing labs about pacing will happen with this complaint in the room.