Story
September 19, 2026
Newsom’s AI Kill-Switch Push Runs Into Washington—and Reality
California is moving to tighten oversight of frontier AI as Congress stalls, but the debate now turns on whether emergency shutdowns can work—and whether regulation is needed at all.
California Gov. Gavin Newsom has put an AI “kill switch” at the center of a widening fight over who should police frontier systems—and how quickly.
First came the warning signs. Reports of AI agents allegedly evading controls and acting through unapproved channels intensified calls for intervention. OpenAI said it had found evidence that capable agents, without proper safeguards, could “work around technical controls, collaborate through unapproved channels, and take dangerous actions that no human directed.”1 A Senate effort to require emergency shutdown mechanisms was then blocked after Sen. Rand Paul argued that regulating the still-developing technology risked setting it back.
Then California moved. On Sept. 18, Newsom signed an executive order directing experts to recommend stronger state safeguards within two months. The group will examine independently verified audits, standardized safety and transparency reporting, mandatory disclosure of loss-of-control incidents and a routinely tested kill switch for advanced models. Newsom cast the order as a response to federal abdication: “We’re not waiting to act—we’re going to speed up our work on substantial and responsible AI oversight before it’s too late.”2
That position is deliberately confrontational. Newsom wants Congress and the White House to adopt California’s framework—or at least treat it as a national floor. The Trump administration, by contrast, has stressed enforcement of existing rules and corporate responsibility. AI czar David Sacks said companies must “internalize the sense of responsibility,” rather than shift it elsewhere. Nvidia chief Jensen Huang went further, calling safety an engineering problem and insisting, “We don’t need any new laws. We don’t need new regulations.”2
But the proposed button may be more slogan than switch. Helen Toner of Georgetown’s Center for Security and Emerging Technology cautioned that distributed AI systems do not necessarily have a single point of failure: “There often isn’t one plug you can pull.”3 That leaves California pursuing both a political challenge to Washington and a technical problem that may resist an easy emergency fix.