California may require advanced AI models to have a kill switch

California is moving its AI policy from voluntary safety guidance toward emergency controls for the most advanced models. Gov. Gavin Newsom issued an executive order on Sept. 18, citing insufficient federal action and directing the state to accelerate oversight, including independent audits and frontier-model safeguards.

The sharpest change is the possibility that developers will have to build an emergency shutoff, or “kill switch,” into advanced AI systems. That would give regulators a direct control mechanism beyond general safety rules, but the order itself does not establish that requirement yet.