The competition among leading AI companies has largely been defined by how quickly they can build and release more capable models. Now, a push by some of those same companies to deliberately slow that progress for safety reasons has created a new legal problem.
Anthropic, OpenAI, SpaceXAI and Google are facing a proposed class action lawsuit that accuses them of violating U.S. antitrust law by agreeing to slow the development of their artificial intelligence systems.
The case was filed on September 18 in the U.S. District Court for the Northern District of California by four consumers who pay for AI subscriptions.
The lawsuit comes less than a week after Anthropic CEO Dario Amodei publicly called for AI companies to reduce the pace at which they improve their most advanced models.
What the Antitrust Lawsuit Alleges
The plaintiffs argue that the companies crossed an important legal line when their leaders publicly supported coordinated efforts to slow AI development.
Their complaint focuses heavily on September 12, when Amodei published an essay titled “We Must Pace the Frontier.” OpenAI CEO Sam Altman, SpaceXAI head Elon Musk, and Google DeepMind co-founder Demis Hassabis later publicly supported parts of his proposal.
The lawsuit claims that the support helped create an agreement among major competitors to restrain the rate at which their products improve.
The case relies on Section 1 of the Sherman Act, which prohibits certain agreements between businesses that restrain competition. The plaintiffs say paid users of Claude, ChatGPT, Grok, and Gemini could receive less value if competing companies collectively decide to improve their products more slowly.
Importantly, the lawsuit does not argue that an individual AI company cannot slow its own development for safety reasons. Its argument is that rival companies cannot privately or collectively replace competition with an agreed pace of development.
However, these are still allegations that have not been proven in court.
Why Amodei Wants AI Development Slowed
Amodei’s proposal is broader than simply delaying new Claude models. He argues that AI capabilities are improving faster than current safety research, testing, and interpretability tools can reliably keep up.
His three-stage plan starts with independent evaluators receiving employee-like access to frontier AI companies so they can review safety practices and investigate incidents, and Anthropic has since committed to implementing this step itself.
Amodei then proposes coordination among U.S. AI companies and eventually international cooperation on controlling the pace of advanced AI development.
Amodei has also warned that increasingly capable autonomous AI agents could become much harder to control. In his essay, he said a more advanced version of recent misaligned AI systems could potentially cause major cyber damage within six to twelve months.
As such, Amodei’s proposal depends, at least in its later stages, on rival AI companies being willing to coordinate around how quickly frontier systems advance. And that is also where the antitrust lawsuit enters the picture.
Safety Coordination Meets Competition Law
The lawsuit puts a legal question around an idea that AI companies have increasingly presented as a safety issue. If the largest developers believe some risks can only be managed collectively, how much can they coordinate before those discussions begin to restrict competition?
Amodei had already acknowledged that concern in his essay. He suggested that government involvement could be necessary for discussions between competing AI companies and raised the possibility of a narrow antitrust waiver for certain safety-related coordination.
The plaintiffs are now challenging that same territory from the other direction. Their case argues that competitors cannot agree to slow the improvement of products that consumers are paying for, even when the stated reason is AI safety.
And this leaves the companies facing two pressures at once. They are being pushed to cooperate more closely on the risks created by increasingly capable AI systems, while competition law limits how far direct rivals can coordinate their commercial behaviour.
Ultimately, the lawsuit could become an early test of how U.S. courts treat industry-wide AI safety agreements when those agreements also affect the pace at which competing products develop.
