Four consumers have sued Anthropic, OpenAI, SpaceXAI and Google, alleging those firms agreed to slow how fast they improve competing consumer AI products.
Filed Friday , the class action landed in the U.S. District Court for the Northern District of California’s San Francisco Division. Charles Buist and Nick Spetsas of Florida, and Cheyenne Hunt and Christine Bullock of California, are the named plaintiffs.
They want to represent a nationwide class of people who bought paid individual subscriptions to ChatGPT, Claude, Grok or Gemini beginning Sept. 12. Relief sought includes treble damages, an injunction and a jury trial.
Rather than independent safety choices, the complaint says, the four companies agreed with rivals to restrain the pace of product improvement.
Public statements starting Sept. 12 sit at the center of that claim. That day, Anthropic CEO Dario Amodei published an essay titled “We Must Pace the Frontier.”
“We must slow the pace at which we improve the capabilities of AI models,” Amodei wrote.
Industry-wide coordination among frontier laboratories, he argued, should set limits on the rate of progress — including training compute, training runs, the use of AI to improve AI, and capability checkpoints.
His proposal also asked the government to issue “a narrow waiver for certain kinds of safety conversations.”
Executives at rival labs soon voiced support.
Within about an hour, Elon Musk — founder and controlling person of SpaceXAI’s Grok business — wrote:
“Dario is right.”
OpenAI CEO Sam Altman also backed the proposal and committed OpenAI to its first step, according to the complaint.
“I agree with Dario that we need to pace the frontier.”
Google DeepMind co-founder Demis Hassabis described the direction as appropriate, while saying implementation still needed work. His remarks were tied to an industry standards body for frontier AI that he had proposed earlier.
Further comments followed. On Sept. 14, Altman said AI progress should not move as quickly as it otherwise could and that OpenAI would begin the work without waiting for an antitrust exemption or new legislation.
A day later, OpenAI Global Policy Chief Chris Lehane said OpenAI had already been working with Anthropic and Google DeepMind on the issue for several weeks.
A working group of representatives from those three companies had formed in July around an industry standards body and had met regularly since then. Plaintiffs treat those earlier contacts as proof that the Sept. 12 posts were not isolated remarks, but public confirmation of talks already underway among competitors.
A July statement on pacing frontier AI development warned of intense competitive pressure against any company that slowed development on its own. It called for an international effort to pace frontier AI on purpose, and for the U.S. government to support that effort.
From that language, plaintiffs infer that the companies understood the commercial cost of going slow alone. Coordination, on their theory, let rivals restrain development without handing an advantage to whoever kept racing ahead.
Section 1 of the Sherman Act is the main legal hook.
An agreement among competing AI companies to cut product quality or slow the rate of improvement is, plaintiffs say, an unlawful restraint of trade — per se, or else under quick-look or rule-of-reason analysis.
Paid subscriptions supply the claimed consumer harm. People subscribe, plaintiffs say, in part to get more capable models and ongoing upgrades. If rivals collectively slow those upgrades, subscribers get less capability or slower improvement for the same price.
Together, the four companies account for at least 80% of paid consumer subscriptions to general-purpose frontier AI assistants in the United States, the suit alleges.
Plaintiffs concede the alleged agreement is recent and that its full effect on released products has not yet shown up. Even so, they allege that investment, training and release decisions have already been affected.
Antitrust risk around a coordinated slowdown is another thread in the filing. It points to a Sept. 10 WIRED report on OpenAI’s questions to members of Congress about whether coordination over an AI slowdown could violate antitrust law.
OpenAI did not dispute that report, according to the lawsuit — a silence plaintiffs call evidence the company knew the legal risk.
Senate bill S. 5105, introduced July 23 by Sen. Adam Schiff, D-Calif., and Sen. Jim Banks, R-Ind., addresses a different problem: foreign theft of AI models.
Under that bill, companies would be exempt from antitrust liability if they coordinate to delay or limit AI development specifically to counter defined security risks — models stolen or weaponized by a foreign nation, for example — and if they notify the Justice Department’s antitrust chief in advance.
That measure predates the Sept. 12 statements by nearly two months.
Google, one of the defendants, is among its endorsers.
“The bill gives the industry clear rules for sharing threat intelligence and collaborating on standards and benchmarks,” Google Vice President John Connell said in July.
Sen. Josh Hawley, R-Mo., raised the exemption issue at a Sept. 16 Senate Judiciary hearing on FBI oversight, while wrapping up questions to FBI Director Kash Patel:
“There is no world in which I will consent to giving the most powerful companies in the history of the world, a small group of three or four of them, antitrust exemptions from our laws so that they can, what? Collude together?”
Hawley did not name the executives.
Competing AI companies reached an agreement to slow development and improvement of their products — that is the core allegation.
“The antitrust laws do not permit competitors to decide among themselves that competition is too dangerous,” the complaint says.
No defendant has filed a response. No court has ruled on whether the alleged agreement existed or violated federal law. The case remains at an early stage.
