Four people who pay for ChatGPT, Claude, Grok or Gemini walked into a federal courthouse on Thursday and accused the four biggest frontier labs of antitrust collusion — not for building AI too fast, but for agreeing in public to build it too slowly.
The complaint, filed on 18 September 2026 in the US District Court for the Northern District of California and reported first by Bloomberg Law, whose account The Next Web reconstructed in detail and which CBS News then syndicated for a general audience, alleges that Anthropic, OpenAI, SpaceXAI and Google violated Section 1 of the Sherman Act when they coordinated to “slow the pace at which their competing frontier AI products improve”. The plaintiffs are Charles Buist and Nick Spetsas of Florida and Cheyenne Hunt and Christine Bullock of California — paying subscribers, suing on behalf of a proposed nationwide class. The case is styled Buist v. Anthropic PBC.
The evidence is the labs’ own announcements
There are no leaked emails in this filing, because there do not need to be. As the complaint itself puts it, the alleged agreement was “proposed in public, accepted in public, and confirmed in public.”
On 12 September, Anthropic chief executive Dario Amodei published “We Must Pace the Frontier”, a three-step plan that included embedded third-party evaluators and industry-wide agreement on limits “on the rate of unchecked AI progress”. According to the filing’s timeline, Elon Musk publicly endorsed the essay within about an hour; Sam Altman wrote that he agreed and committed OpenAI to the plan’s first step; and Google DeepMind’s Demis Hassabis publicly backed the essay’s direction. Two days later, Altman said OpenAI would not wait for an antitrust exemption before starting the work, and on 15 September OpenAI’s policy chief Chris Lehane confirmed the company had been working with Anthropic and Google DeepMind on the issues for several weeks.
The complaint treats that sequence as the whole case: an offer, four acceptances, and two confirmations of implementation. OpenAI had even asked Congress whether coordinating a slowdown would be legal — and the filing cites that question as evidence the companies understood the conduct as coordination that needed a legal shield.
What the plaintiffs are not arguing
The filing is narrower than the headlines suggest, and both sides of the AI safety argument should read it carefully. The plaintiffs state in the complaint that they take AI risk seriously and want guardrails — set “by we the people, through state and federal regulation, and through the voices of juries”, not by four companies. They expressly do not challenge unilateral slowing, independent safety evaluators, or any company’s own decisions about its pace. Their target is the coordination, not the caution.
“The antitrust laws do not permit competitors to decide among themselves that competition is too dangerous,” the plaintiffs argue — the judgment about whether competition is too dangerous belongs to Congress, not to the market participants themselves.
Lead counsel Nick Rowley put it more bluntly in a statement: AI safety decided “by private self-serving agreements between the world’s most powerful ‘for profit’ technology companies” is the thing the suit exists to stop.
The weaknesses are built into the filing
Buist v. Anthropic is, so far, a case with no harm to point at. The complaint concedes the full effect on released products “has not yet manifested” — no slowed release, cancelled model or altered training run is identified anywhere in its 29 pages, as The Next Web’s analysis noted. The claimed 80% share of the US paid-subscription market is pleaded “on information and belief” — a legal phrase meaning the plaintiffs believe it is true but are not yet swearing they measured it. Musk’s “acceptance” was four words on social media, and Google’s endorsement left implementation details open.
Courts have historically been skeptical of antitrust claims built on public statements and inferred agreements rather than direct evidence of collusion. The defendants will argue that discussing industry standards is normal and lawful; the plaintiffs will counter that agreeing to limit capability improvements for paying customers is a different thing entirely.
Anthropic, OpenAI, Google and SpaceXAI had not responded to the filing as of the weekend coverage — CBS noted representatives did not immediately respond to a request for comment on Saturday. None has publicly answered the complaint as of publication.
Why the case cuts at the labs’ own framing
The lawsuit lands on the exact fault line the pacing debate has been walking since Amodei’s essay: if a slowdown is genuinely safe and necessary, each lab can do it alone — and Anthropic, for one, has already committed unilaterally to embedded third-party evaluators. If coordination is genuinely necessary, then the safety rationale and the antitrust exposure are the same fact wearing two hats. Amodei anticipated this in his essay, writing that the US government would need to “issue a narrow waiver for certain kinds of safety conversations” for cross-lab coordination to be lawful. The safe-harbour bill that would have provided one, the Collaboration on Adversarial Threats and Security Risks Act, remains unenacted.
The complaint’s answer to the labs’ dilemma is the courts’: if coordination to slow AI is what the companies want, the plaintiffs argue, they should have sought the exemption — not proceeded anyway. Meanwhile the geopolitical frame keeps tightening: Beijing’s state media has already called the entire slowdown push a “Cold War script”, and the EU has invited the labs to Brussels to discuss pacing, meaning the four defendants now face a collusion allegation in one jurisdiction and an invitation to coordinate in another.
The suit also creates a strange alignment of interests. The plaintiffs say they are not anti-safety; they are anti-cartel. The labs say they are not anti-safety; they are just waiting for a legal safe harbour to coordinate on it. Both sides are asking the same government to act — one to permit the coordination, the other to forbid it — and the same filing is now evidence in both arguments.
❓ FAQ
Is this an actual antitrust case or a publicity stunt?
It is a real class-action complaint filed on 18 September 2026 in the Northern District of California, obtained and reported by Bloomberg Law. That said, it is at the allegation stage: no defendant has responded, and the complaint identifies no released product that was actually slowed. Courts will decide what it is worth.
Did the four companies really agree to slow AI down?
That is the plaintiffs’ characterisation of a sequence of public statements on 12-15 September, in which Amodei proposed industry-wide pacing and Musk, Altman and Hassabis publicly endorsed it. The companies have not publicly conceded any agreement exists, and the complaint cites their public statements — not internal documents — as its evidence.
Can companies be liable for agreeing to make their products safer?
The plaintiffs’ theory is that agreeing to make products worse is an output restriction regardless of motive — and that safety can be pursued unilaterally, through independent evaluators, or through regulation instead. Sherman Act jurisprudence has no exemption for good intentions; but no court has yet said whether “slowing capability growth” is the kind of restraint Section 1 condemns.
What do the plaintiffs want?
Treble damages for the proposed class of paid subscribers and an injunction stopping the alleged coordinated slowdown, per the filing. The complaint does not quantify damages at this stage.
📰 Sources
- CBS News — Lawsuit says AI giants made illegal deal on technology’s slowdown (19 September 2026)
- The Next Web — Buist v. Anthropic: antitrust, waiver and the evidence of a public agreement (20 September 2026)
- Unite.AI — Consumers sue Anthropic, OpenAI, SpaceXAI and Google over alleged AI pact (18 September 2026)
- Yahoo Finance — Four paid subscribers are suing the biggest AI labs for coordinating
