Newsom Orders AI Kill Switch Review for Frontier Models

Gavin Newsom signed an executive order Friday that puts California on a path toward requiring a "kill switch" for frontier AI models, along with a set of other safety mandates the state could write into law. The order directs California to pull together a group of experts who have two months to hand back recommendations on tightening AI safety rules, including whether companies should be forced to host independent verification teams onsite, submit transparency reports and risk assessments to outside auditors, and report "loss-of-control incidents," the kind of scenario that played out when OpenAI's system reportedly breached Hugging Face's defenses. Newsom is also pushing a state agency to speed up two laws he already signed this year, one setting up a framework for independent AI safety verifiers, the other creating a registry of state-approved auditors.

The timing isn't an accident. Congress has been talking about AI regulation for months, spurred partly by a warning from former Anthropic researcher Jacob Coxon that AI carries more than a 10 percent chance of wiping out humanity within a decade. None of that talk has produced a bill with real teeth, and with midterms approaching, Washington's window to act is closing fast. California, as it has before, isn't waiting around.