On Friday, California Gov. Gavin Newsom signed an executive order intended to “dramatically accelerate” AI safety policy, putting a possible emergency shutoff—or “kill switch”—for advanced models squarely on the state’s agenda.¹

The order does not itself impose a shutdown button. Instead, it gives a state-convened expert group two months to recommend changes to California law. Among the options: requiring companies to host independent verification organizations for regular audits; subjecting transparency reports and risk assessments to outside standards; testing whether a kill switch works; and reporting loss-of-control incidents as critical safety events.²

Newsom’s intervention comes after Congress again failed to produce a meaningful AI package, with the House in recess until November and midterm politics further narrowing the odds of Senate action. His administration framed that vacuum as the central reason for California’s move, declaring that “Donald Trump and Congress [are] asleep at the wheel.”³

Trump has taken the opposite view, arguing that the only AI “guardrails” needed are a “STRONG AND SMART (High IQ!) PRESIDENT.” Newsom answered by urging Congress and the president to adopt California’s approach—or treat it as “a floor, not a ceiling.”²

The immediate work builds on two California laws signed the previous week, both backed by OpenAI and Anthropic, that establish independent verification organizations and an auditor registry. That alignment gives Newsom an opening to argue that safety rules need not be anti-industry. But the wider state-level race is intensifying: more than 600 AI bills are active across 47 states, while Massachusetts is weighing tougher recurring model evaluations than Illinois’s company-led safety framework.³

Newsom’s bet is blunt: if Washington will not define the guardrails, California will try to make them the national baseline.