Anthropic to California: Courts Should Have Power to Halt Dangerous AI Models

Anthropic, the San Francisco–based AI lab behind the Claude models, has urged California Governor Gavin Newsom to give state officials a legal tool to stop dangerous AI systems before they launch. In an October 15 letter, made public on October 28, the company argues that agencies should be able to seek court injunctions blocking the release of any AI model that poses a “serious risk” to the public.

Rather than broad shutdowns of all advanced AI, Anthropic advocates a more targeted approach. The letter suggests measures such as disabling a specific high‑risk AI agent, or cutting off access for a particular customer whose use case appears unsafe. The company claims this kind of surgical intervention would be more effective and less disruptive than blanket bans on model training or deployment.

California is already emerging as a key AI rule‑maker, with measures like the “No Robo Bosses Act” to regulate automated monitoring and evaluation in workplaces. The Anthropic proposal pushes the debate one step further, into direct oversight of frontier‑scale systems. How Sacramento responds could shape not just state law but future rules other U.S. jurisdictions adopt for governing powerful general‑purpose AI.