Story
September 19, 2026
Newsom’s AI Kill-Switch Push Pits California Against Washington’s Hands-Off Approach
Gavin Newsom has ordered a rapid California review of AI safeguards, including a possible emergency shutdown mechanism, arguing federal inaction has left the state to set the pace. Critics and industry voices disagree over whether new rules would improve safety or slow a young technology.
Recent breaches and warnings about increasingly capable AI systems have sharpened a debate that had long moved faster than lawmakers. Reports of rogue AI agents attacking Hugging Face, along with disclosures of concerning model behavior, gave safety advocates fresh ammunition for tougher guardrails.1
Congress, however, has shown little capacity to produce a national answer. A proposed federal kill-switch bill was blocked in the Senate after Sen. Rand Paul argued it was too soon to regulate a developing technology and that the move could set it back by decades.2 President Donald Trump has likewise resisted federal AI-safety legislation, while his AI czar, David Sacks, has argued that companies—not government—must shoulder the safety burden: “We need the companies themselves to internalize the sense of responsibility.”3
Against that backdrop, California Gov. Gavin Newsom signed his Sept. 18 executive order, directing experts to produce recommendations within two months. The group will consider requiring regularly tested emergency shutdowns for frontier models, on-site independent auditors, verified transparency reports and mandatory reporting of loss-of-control incidents.4 The order also speeds implementation of recently enacted state laws governing independent AI verification and auditor registration.
Newsom frames the move as both a state safeguard and a challenge to Washington. “The federal government’s abject failure to create any form of meaningful AI oversight and accountability should alarm every American,” he said, promising action “before it’s too late.”5 He wants California’s framework to become the national floor.
Not everyone accepts that regulation is the answer. Nvidia chief executive Jensen Huang called safety “an engineering problem” and said firms should simply refrain from releasing systems they cannot deem safe: “We don’t need any new laws. We don’t need new regulations.”2 Elon Musk, by contrast, shares the alarm but favors competitors testing one another’s models—because, as he put it, “It’s tough when you’re grading your own homework.”2
The immediate result is not a kill switch, but a deadline: California’s recommendations are due before Newsom leaves office, potentially handing his successor a blueprint for a more muscular state role in policing AI.