Four of the industry’s leading AI labs: Anthropic, OpenAI, Elon Musk’s xAI (referred to in the filing as SpaceXAI), and Google, are facing a proposed federal class action accusing them of illegally coordinating to slow the pace of AI model improvement, a move the plaintiffs say violates Section 1 of the Sherman Act.
The lawsuit, filed in the U.S. District Court for the Northern District of California, was brought by four consumers who pay for Claude, ChatGPT, Grok, or Gemini subscriptions. They aim to represent a nationwide class of other paying users, arguing that when rival labs jointly agree to advance more slowly than each would on its own, the result is a restricted, inferior product for consumers — precisely the kind of collective restraint antitrust law prohibits.
How the alleged pact came together
According to the complaint, the pact became public on September 12, 2026, when Anthropic CEO Dario Amodei published an essay titled “We Must Pace the Frontier,” arguing that labs should jointly set limits on the speed of unchecked AI progress so no single company loses ground by moderating alone. Within hours, xAI’s Elon Musk, OpenAI’s Sam Altman, and Google DeepMind co-founder Demis Hassabis each voiced support. Musk posted that “Dario is right,” Altman committed OpenAI to an early step of the plan, and Hassabis called the essay “the right path forward.”
The lawsuit contends the coordination didn’t start there. It points to a July 2026 open letter, signed by senior figures across several frontier labs, that acknowledged competitive pressure discourages any single company from slowing down alone and called on governments to help build mechanisms for a broader, coordinated slowdown. The complaint treats that letter, along with subsequent working-group meetings among Anthropic, OpenAI, and Google, as the actual start of the alleged coordination — roughly two months before Amodei’s essay.
What plaintiffs are and aren’t challenging
The plaintiffs say they aren’t objecting to any single company’s independent decision to invest more in safety or release models more cautiously. Their target is the alleged substitution of mutual restraint for individual competitive accountability: a lab that slows down alone risks losing users, talent, and technical edge, but an understanding that rivals will all ease off together removes that penalty — which plaintiffs argue functions as an unlawful agreement to limit output. They’re seeking an injunction, a declaration that the pact is unlawful, and damages.
Meta’s pointed absence
Meta, notably, is not named as a defendant. CEO Mark Zuckerberg has been the industry’s most vocal critic of the pacing push, pushing back directly on Amodei’s essay and arguing each lab has its own incentive and responsibility to train models safely without a coordinated slowdown. He has pointed to Meta’s own multi-month delay of its Muse AI agent. It was undertaken, he said, to strengthen its safety and security, as an example of unilateral caution, while not calling on rivals to follow suit. At the same time, after the agent’s launch, Amazon, for example is forced to block Muse from accessing its platform since it doesn’t have an opt-in function and appears to capture and store customer credentials. Meta’s chief AI scientist had earlier signed the July letter, even as Zuckerberg publicly dismissed much of the industry’s “doom” talk.
The defendants have not yet issued detailed public responses. Courts will now have to decide whether the public statements, prior joint letter, and reported working-group contacts amount to an illegal combination or remain protected speech and advocacy.
