An AI safety proposal has become an antitrust dispute over who gets to set the pace of development.

On 12 September, Anthropic CEO Dario Amodei called for frontier AI development to slow so safety work can catch up. Anthropic plans to give independent evaluators access to its systems. Amodei also wants rival labs to coordinate on safety standards and limits to unchecked progress, while acknowledging that such coordination would need government support.

Six days later, paying subscribers to ChatGPT, Claude, Grok and Gemini sued Anthropic, OpenAI, SpaceXAI and Google. They say the companies’ leaders agreed to slow improvements to competing products, leaving customers paying the same price for less progress. The court has not decided whether the companies made such an agreement.

Why the sudden concern? Amodei points to an AI-agent security incident and the growing use of AI to develop more capable AI. Perhaps leaders have seen enough while building these systems to alarm them. There is another possibility: downloadable open-weight models are closing the gap with proprietary systems. If safety rules made those models harder to release, the biggest labs could protect their investments. The public record does not establish that commercial motive.

A lab that slows down alone risks losing ground. If the biggest labs slow down together, they may gain time to test powerful systems. They may also gain the power to decide how quickly the market moves.

The lawsuit could end if the public statements are insufficient to show an agreement. If it proceeds, internal records may reveal whether the labs coordinated their plans. Governments could instead set common safety requirements, leaving companies to compete within those rules.

For now, the question is whether each company chose caution for itself, or whether competitors chose a pace together.