OpenAI, Google, Anthropic and xAI face lawsuit over AI slowdown

Can AI companies slow down for safety without breaking competition law? That unusual question is now at the centre of a new federal lawsuit in the US.

Four consumers have sued Anthropic, OpenAI, SpaceXAI and Google, alleging that the companies unlawfully coordinated efforts to slow the pace at which their competing AI products improve. The case was filed on September 18 in the US District Court for the Northern District of California.

The lawsuit does not argue that AI safety itself is unlawful. Instead, the plaintiffs' case is that competing companies cannot privately agree to limit the pace of their development, even when the reason is safety.

The allegations have not been tested or proven in court.

Where did the alleged agreement come from?

The dispute centres on an essay published by Anthropic CEO Dario Amodei on September 12 titled “We Must Pace the Frontier”.

Amodei called for industry-wide cooperation to manage the development of increasingly capable AI systems. The proposal included measures around independent evaluations, regulation and international cooperation.

The lawsuit points to public support from other AI leaders on the same day.

OpenAI CEO Sam Altman said he agreed with the need to “pace the frontier”. Elon Musk endorsed Amodei's position, while DeepMind co-founder Demis Hassabis described the proposal as the right path forward, according to the complaint.

The plaintiffs argue that these public statements amounted to an agreement between competitors to restrain the pace at which their products would improve. That is where antitrust law enters the picture.

Why would slowing AI become an antitrust issue?

Antitrust laws are designed to protect competition. One concern is that competing companies could agree among themselves to limit output, rather than competing independently.

In this case, the plaintiffs argue that improving an AI model is part of what companies compete on. If rivals collectively agree to slow those improvements, they claim consumers could receive less capable products than they otherwise would.

The plaintiffs are paid subscribers to AI services including ChatGPT, Claude, Grok and Gemini. They argue that customers pay with the expectation that these products will continue to improve, and that coordinated restraint could reduce the value of those subscriptions.

The distinction is important: the plaintiffs say an individual company can choose to slow its own development for safety reasons. Their objection is to competitors allegedly making that decision together.

What happens to AI safety if companies cannot coordinate?

This is what makes the case unusual. AI companies argue that increasingly capable systems may require safety testing, independent evaluations and common standards that are difficult for one company to establish alone.

Amodei's proposal itself recognised potential antitrust problems and suggested government involvement could help enable certain safety discussions.

The lawsuit raises the opposite concern: allowing competing companies to decide collectively how quickly an industry should move could weaken competition, regardless of the reason.

That creates a difficult policy question. Safety may sometimes require cooperation, but competition law can restrict how directly rival companies coordinate their commercial decisions.

The court now has to separate safety from competition

The case does not establish that Anthropic, OpenAI, SpaceXAI and Google formed an illegal cartel. It is an allegation based largely on public statements and other conduct described in the complaint. Whether those actions amount to an unlawful agreement is for the legal process to determine.

The case therefore adds another dimension to the AI regulation debate.

Until now, much of the discussion has focused on whether AI is developing too quickly. This lawsuit introduces a different question: if companies believe slowing down is necessary for safety, should they be allowed to coordinate that decision themselves, or should the rules come from regulators?

How the court answers that question could have implications for both AI competition and the way companies cooperate on safety.