Sued for Agreeing to Slow Down: Four Subscribers Take the Pace-the-Frontier Pact to Federal Court

On the evening of Friday, September 18, 2026, six days after Anthropic CEO Dario Amodei published “We Must Pace the Frontier” and Elon Musk, Sam Altman and Demis Hassabis publicly agreed with it, four paying subscribers sued all four companies — for agreeing. The proposed class action, filed in the U.S. District Court for the Northern District of California in San Francisco, is the first lawsuit to treat the AI industry’s slowdown coordination not as a safety story but as an antitrust one. The case is Buist v. Anthropic PBC, No. 3:26-cv-10693.

The irony is the point. For most of 2026, the loudest arguments in AI have been about whether the frontier labs are moving too fast. This suit argues that, the moment the four most powerful labs began coordinating how fast their products improve, they crossed a line that has nothing to do with safety at all: Section 1 of the Sherman Act, the 136-year-old law that forbids agreements among competitors that restrain trade.

Who is suing, and what exactly they claim

The plaintiffs are four consumers who pay for the products in question. Charles Buist and Nick Spetsas are attorneys in Florida; Cheyenne Hunt is a lawyer in California, and Christine Bullock, also of California, is a Claude subscriber. Buist, Hunt and Spetsas each say they pay for all four services — ChatGPT, Claude, Grok and Gemini. They are represented by Nicholas C. Rowley and Trial Lawyers for Justice, and they demand a jury.

The single count is Section 1 of the Sherman Act, pleaded in three alternative ways: as per se illegal, and in the alternative under “quick-look” and rule-of-reason analysis. The framing, as quoted by Bloomberg Law from the filing, is deliberately narrow: “Plaintiffs challenge only what the antitrust laws forbid: an agreement among competitors about how fast their competing products will improve.” And, on the question of whether safety motives provide cover: “Congress has granted no exemption for that agreement.”

The alleged market is specific: paid consumer subscriptions to general-purpose frontier assistants — the paid tiers of ChatGPT, Claude, Grok and Gemini — which the four defendants are alleged, on information and belief, to control at least 80 percent of. The injury is similarly concrete: the same roughly $20 a month for products that improve more slowly than competition would produce. If the class is certified and the plaintiffs prevail, the Clayton Act’s treble-damages provision would multiply that injury across every paying subscriber in the country.

The chain the complaint draws

The complaint’s story does not start with Amodei’s September 12 essay. According to Unite.AI’s reading of the filing, it starts two months earlier:

• July 2026 — Representatives of Anthropic, OpenAI and Google form a working group that meets regularly on an industry standards body.
• July 14 — Google DeepMind’s Demis Hassabis publicly proposes a U.S.-led frontier-AI standards body, modeled in part on the Financial Industry Regulatory Authority (FINRA).
• July (continued) — A statement titled “Pacing the Frontier,” organized with Guidelight AI Standards and Encode AI, gathers 1,386 signatories from frontier labs, including Amodei, Jared Kaplan, Jack Clark, Jakub Pachocki, Mark Chen and Shane Legg, asking the government to support a deliberate pacing effort.
• September 6 — OpenAI publishes chief scientist Jakub Pachocki’s essay “An Alien Mind,” naming coordination among developers to slow development as a principal option.
• September 10 — WIRED reports that OpenAI privately asked members of Congress whether a coordinated slowdown could violate antitrust law.
• September 11 — Altman tells Fortune a common industry plan will happen, declining to detail private talks.
• September 12 — Amodei publishes “We Must Pace the Frontier.” Musk agrees within an hour; Altman commits OpenAI to “step one”; Hassabis calls it the right path forward.
• September 14 — Altman says progress “will proceed more slowly than it could” and that OpenAI will not wait for an antitrust exemption or legislation to begin the work.
• September 15 — OpenAI’s Chris Lehane confirms the company has been in talks with Anthropic and Google DeepMind “for several weeks.”
• September 18 — The complaint is filed in San Francisco. Hunt announced it on X at 00:23 UTC on the 19th.

Read as a whole, the timeline is the lawsuit. Each public statement that safety observers treated as progress — the essay, the endorsements, the confirmed talks — becomes, in the complaint, an exhibit of the agreement being executed.

The theory: slower is less

The economics of the claim are the economics of any output cartel, applied to software that improves instead of a good that ships. Subscriptions are marketed and priced on access to each company’s most capable models and their continuing improvement. An agreement that slows improvement therefore lowers what subscribers get for the same price. The line the plaintiffs use to explain why an agreement was needed at all is one an antitrust court will hear as the heart of the case: “A company that slows alone while rivals race ahead loses customers, revenue, talent, and technological leadership.”

That sentence is also, almost word for word, the argument the essay itself made for why the industry could not slow down one company at a time. The plaintiffs’ innovation is to take the labs’ own public reasoning and hand it to a Sherman Act framework.

The plaintiffs, in their own words

Lead counsel Rowley, in a statement carried by Politico, framed the case as being about who sets the rules, not whether rules are needed: “Humanity deserves iron clad safeguards when it comes to extinction event threats such as nuclear warfare and now the biggest risk to mankind in history.” And: “The rule of law should be established transparently and lawfully by our government, with accountability to the public.”

Hunt was blunter on X: the plaintiffs are suing because “we deserve real AI safety standards, not shady deals cut by an unaccountable cartel of billionaires in a back room.” Her post also argues the September 12 announcement gave the House cover to adjourn without passing binding legislation — a claim that is hers, not the complaint’s as reported, and offered without a source.

Nobody was surprised by the word “antitrust”

What makes the suit credible rather than eccentric is that the antitrust question was on the record all week before anyone sued. Amodei’s essay itself footnoted the need for government mediation or antitrust waivers for its second step. Former FTC chair Lina Khan said on September 13 there is no AI exemption from laws already on the books. Nvidia’s Jensen Huang told CNBC a waiver was “completely unnecessary.” OpenAI said it would not wait for one.

In other words: everyone involved had already conceded that the legality of coordinated pacing was an open question. The plaintiffs’ move was simply to walk it into a courthouse.

What to watch

Three things now matter. First, whether any of the four defendants responds — at publication time, none had responded to requests for comment from Politico or Bloomberg Law. Second, whether the court entertains the per se theory; most observers expect the fight to center on whether a “quick look” or full rule-of-reason analysis applies to agreements about the rate of improvement of a service, a genuinely novel question. Third, the tension the suit will live inside: on the same night the complaint surfaced, Reuters reported — citing three sources — that Anthropic is weighing a new model to counter OpenAI’s momentum since GPT-6 Astra, ahead of an expected November IPO. A lab weighing a counter-launch while being sued for agreeing to slow down is the contradiction this case was built to expose.

The deeper stakes cut both ways. If the suit survives a motion to dismiss, every future coordination effort among the labs — the standards body, the shared safety bars, the information exchanges — will need antitrust counsel in the room, and some of it may simply stop. If it fails, the industry gets a de facto green light to coordinate pacing without statutory cover, and the safety argument wins by default. Either way, the question of how fast AI should improve — until now a debate among executives, researchers and regulators — now has a docket number.