Anthropic urges California to block unsafe AI models via court orders

The letter supports regular risk reporting—at least every six months—for large AI developers and independent oversight to prevent harm before it occurs. Anthropic opposes blanket kill-switch rules, instead advocating for targeted interventions like disabling specific agents or restricting access for problematic users. The company’s stance contrasts with broader legislative efforts, such as the AI Kill Switch Act debated in Congress, which would mandate shutdown mechanisms for high-risk models. Anthropic’s recommendations may influence California’s approach and broader US policy discussions.