California is drawing up plans to switch off AI. Governor Gavin Newsom has issued an executive order asking state agencies to recommend stronger independent reviews of the technology and to consider a shut-off function should it get out of control, while a convened panel of experts has two months to draft guidelines for the state's next AI safety laws. To be clear about scale: the order does not create a kill switch. It starts the machinery that could, and it puts the idea on the desk of the government that regulates most of the industry's builders.
Watch Desk analysis
What happened
Per the governor's office, the executive order asks state agencies to recommend legislative changes that would enhance independent reviews of AI, and to weigh a shut-off function for systems that go rogue. It also convenes a group described as world-leading experts to write guidelines within two months, material California can use to develop further safety laws.
Who those experts are is not yet public. The order names nobody, though Newsom has previously praised researchers at the University of California, MIT and Carnegie Mellon, along with Stanford's Fei-Fei Li, "godmother of AI" title and all.
The timing is no accident. This month, Jacob Coxon, a former researcher at Anthropic and OpenAI, quit over fears that unchecked AI could wipe out civilisation, and Bilal Chughtai, once a safety researcher at Google DeepMind, wrote that he earnestly believes AI has the potential to kill us all. Newsom had already signed a stack of technology bills on 10 September while calling AI's trajectory a "five-alarm fire".
There is history here too. In 2024, Newsom vetoed Senate Bill 1047, Senator Scott Wiener's attempt to add safety, testing and liability requirements to firms such as Anthropic and OpenAI, which carried its own kill-switch clause. He signed a slimmer successor, SB 53, in 2025. Wiener is delighted this time, telling the New York Post: "Requiring emergency shutdown procedures, or kill switches, provides tangible guardrails to the public to prevent these disasters."
What changes
- Independent reviews on the table
Agencies are asked to recommend legislative changes that strengthen independent reviews of AI. - A shut-off function under study
Agencies are to consider how a kill switch would work for AI that gets out of control. - Expert guidelines in two months
A convened expert group must draft guidelines to shape California's next safety laws. - The advisers are unnamed
No experts are listed yet, so who shapes the guidelines is still an open question.
Why it matters
California is not just any state in this argument. Most frontier labs build there, which means state rules work as near-national rules for the industry, and a governor who vetoed the defining AI bill of 2024 is now asking his own agencies to sketch a shut-off mechanism. The kill-switch debate has spent the week being told, by a blocked Senate bill and sceptical AI leaders, that a button is harder than it sounds. The most consequential state government in AI has commissioned the homework anyway.
The two-month clock is the part with teeth. Guidelines drafted by named experts, feeding into legislation, is a path from warnings to bill text, and it lands before the year is out.
Our read
Treat this as a commission, not a law. Executive orders are to-do lists with letterheads, and the order's own softness, unnamed experts and a "consider" on the kill switch, is the honest state of play. What earns your attention is the deadline: if the expert guidelines name testable shutdown mechanisms, California will have moved the argument further than Congress managed this week. If you build or deploy models in California, read those guidelines when they land, because they are a draft of your next compliance problem.
What to watch
- The expert list, and whether it includes names the labs trust or names the labs dread.
- Whether agency recommendations describe a verifiable shut-off standard or merely study one.
- Any bill text reviving SB 1047's kill-switch clause with the governor's blessing.
- The labs' response, given Anthropic and OpenAI were named in the 2024 fight.
Discussion spark: Newsom vetoed the kill-switch bill in 2024 and the Senate blocked its own version this week: if California's agencies actually design a shut-off, what can a state switch off that companies could not simply relocate away from?
Sources and evidence
- Gavin Newsom plots ‘kill switch’ if AI gets out of control – New York Post (18 September 2026, 15:37 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.