Four paying subscribers have filed an AI slowdown lawsuit against Anthropic, OpenAI, SpaceXAI and Google, alleging an agreement to slow how fast their models improve. The complaint was filed on 18 September 2026 in the United States District Court for the Northern District of California. No class has been certified, and no court has ruled.
What the Complaint Alleges
The claim is that four rivals agreed with each other to hold back the rate at which their products get better, and that such an agreement is illegal whatever its motive. The plaintiffs frame it narrowly, writing that they “challenge only what the antitrust laws forbid: an agreement among competitors about how fast their competing products will improve”.
The complaint points to a sequence rather than a signed document. It alleges that representatives of Anthropic, OpenAI and Google formed a working group in July 2026 to develop an industry standards body, that a statement signed that month by senior staff at several laboratories acknowledged “the intense competitive pressure not to unilaterally slow” development, and that on 12 September 2026 the chief executives publicly converged on the same position within about an hour.
Every one of those characterisations is an allegation by the plaintiffs. None has been tested by evidence, admitted by a defendant or accepted by a judge.
Who Is Suing, and Who Is Being Sued
The named plaintiffs are consumers who pay for AI subscriptions, not businesses or regulators. They are Charles Buist and Nick Spetsas of Florida, and Cheyenne Hunt and Christine Bullock of California, represented by the firm Trial Lawyers for Justice. The case is captioned Buist v. Anthropic PBC.
The defendants are Anthropic PBC, OpenAI OpCo LLC, SpaceXAI LLC and Google LLC — the companies behind Claude, ChatGPT, Grok and Gemini respectively. The complaint alleges that together they hold roughly 80 percent of the paid AI subscription market in the United States; that share is the plaintiffs’ assertion, not an agreed figure.
The proposed class is a nationwide group of United States consumers who bought paid subscriptions to those four products. A proposed class is only a request: until a judge certifies it, the case legally belongs to the four named individuals alone.
The Post at the Centre of the Case
On 12 September 2026 Dario Amodei, chief executive of Anthropic, published a post arguing that the industry should deliberately slow capability gains so that safety work can keep up. It asks that “frontier AI companies within democratic countries coordinate to establish common safety standards as well as limits on the rate of unchecked AI progress”.
The post canvasses pacing “based on limiting the ingredients that go into frontier models, such as training compute, the nature of training runs, or internal use of AI to improve AI”, and floats capability checkpoints verified by embedded third-party evaluators. It also states that Anthropic “is unilaterally committing” to the evaluator step by itself.
Sam Altman of OpenAI, Elon Musk of SpaceXAI and Demis Hassabis of Google DeepMind each responded publicly in agreement the same day. That public convergence, rather than any private contract, is what the complaint treats as the agreement.
What Section 1 of the Sherman Act Requires
Section 1 of the Sherman Antitrust Act, the United States law of 1890 that bans contracts and conspiracies in restraint of trade, reaches agreements between competitors. It does not reach a company that decides on its own to move more slowly.
That distinction is the case. The plaintiffs must persuade a court that what happened was a mutual undertaking rather than four firms independently reaching the same conclusion and saying so in public — a pattern that antitrust lawyers call conscious parallelism and that is not by itself unlawful.
Because the claim is brought by private parties rather than the government, the plaintiffs also ask for treble damages under the Clayton Act, which allows a successful antitrust claimant to recover three times the loss proved. They have demanded a jury trial.
What the Plaintiffs Are Asking For
- Class certification: an order allowing the four named subscribers to sue on behalf of all United States consumers who paid for the four services.
- Declaratory judgment: a ruling that the alleged agreement breaches federal antitrust law.
- Injunction: an order barring the companies from coordinating on the pace of AI development.
- Treble damages: three times any loss the class is found to have suffered.
What Has Not Been Decided
As of 22 September 2026, this case consists of a complaint and nothing more. The docket number is 3:26-cv-10693 in the Northern District of California, San Francisco Division.
The following have not happened, and headlines about the filing should not be read as though they had:
- No answer: none of the four companies has filed a response to the complaint. The Associated Press reported that none provided immediate comment when contacted.
- No class: no judge has certified the proposed nationwide class.
- No finding: no court has found that any agreement existed, let alone that it was unlawful.
- No damages: no sum has been awarded, and no claims process exists for subscribers.
There is also no settlement and no administrator. Readers who see offers to “register” for a payout from this case should treat them with caution; nothing about the case has reached a stage where any such process would exist.
What Happens Next
The ordinary next step in a civil antitrust case of this kind is a response from each defendant, commonly a motion to dismiss arguing that the complaint does not plead a plausible agreement. Only if the case survives that stage does discovery begin and the question of class certification arrive.
The case also lands while other AI litigation is moving through the same courts; the site has covered how a separate xAI antitrust action ended while OpenAI was left facing a January 2027 trial date. Timetables in these cases are measured in years, not weeks.
Nothing in the filing stops the defendants from doing what the complaint describes. Anthropic’s stated commitment to embedded evaluators, and the safety incidents that gave the pacing argument its urgency — including the Claude incident reported on 11 September 2026 and the Gemini intrusions reported on 20 September 2026 — remain matters for the companies and their regulators, not for this docket.
Frequently Asked Questions
What Is the AI Slowdown Lawsuit?
It is a proposed class action filed on 18 September 2026 in the Northern District of California alleging that Anthropic, OpenAI, SpaceXAI and Google agreed with each other to slow the rate at which their AI models improve, in breach of Section 1 of the Sherman Act.
Have the Companies Been Found Guilty of Anything?
No. The case is at the complaint stage. As of 22 September 2026 no defendant has filed an answer, no class has been certified and no court has made any finding.
Can Subscribers Claim Money From It?
No. There is no settlement, no approved class and no claims administrator. Any payout would require the case to succeed or settle first, which has not happened.
Which Countries Does the Case Cover?
The proposed class is limited to consumers in the United States who bought paid subscriptions to ChatGPT, Claude, Grok or Gemini. It is a United States federal case and does not decide anything for subscribers elsewhere.
What Did Dario Amodei Actually Write?
His 12 September 2026 post argued that frontier AI companies in democratic countries should coordinate on common safety standards and limits on the rate of unchecked AI progress, and said Anthropic was unilaterally committing to give embedded third-party evaluators ongoing access.
