Newsom gave experts two months to recommend an AI kill switch and auditors inside frontier labs
The embedded evaluators Dario Amodei offered to volunteer for six days earlier are now on the list of things California is working out how to require.
On September 18, 2026, Gavin Newsom signed an executive order telling California's Government Operations Agency to speed up two AI oversight laws he signed the week before, and to bring in outside experts on what the state should require next. They get two months.
Four items sit on the list those experts have to work through. One is an emergency shutoff for frontier models, which the governor's office calls a "kill switch." Another is outside auditors working onsite in the labs.
- Signed
- September 18, 2026
- Who
- Governor Gavin Newsom
- Experts report back
- Within two months
- Laws accelerated
- SB 813 and AB 1405
- Also on the list
- An emergency shutoff for frontier models
That second one surprised me. On September 12, 2026, Dario Amodei asked the industry to slow down and offered embedded evaluators as something Anthropic would do on its own, with no law behind it (METR is the example he gives). Six days later California started working out how to require it.
What the order tells the state to do
Two laws get moved up. SB 813 sets up a framework for certifying independent verification organizations, and AB 1405 creates a state registry of AI auditors with standards for their independence, transparency and integrity. Both were signed last week, so neither is fully running yet.
The rest needs new law. Newsom tells the Government Operations Agency, working with the Governor's Office of Emergency Services, to convene national experts and come back with recommendations for changing it. The first of them reads like the essay Amodei published six days earlier.
Require frontier AI companies to embed a designated independent verification organization onsite in their labs to conduct regular audits and evaluations.
Number two would take the safety frameworks, transparency reports and risk assessments that frontier companies already file under state law, and have an independent verification organization check them against standards it considers adequate. So filing wouldn't be enough on its own.
The kill switch, and who checks it
Third on the list is the shutoff itself, and the order doesn't stop at asking for one.
Advance the creation of a “kill switch” for frontier models, with the efficacy of the switch verified on an ongoing basis by an independent verification organization.
Number four would widen what state law counts as a critical safety incident to cover loss-of-control incidents, and it names one: the Hugging Face attack. (Amodei's essay describes the same incident.) SB 53, the 2025 Transparency in Frontier Artificial Intelligence Act (the release calls it the nation's first AI safety law), already makes frontier developers report specified critical safety incidents to the state and protect whistleblowers who report serious risks.
What he wants Washington to do
About as much of the release goes on the federal government as on the order.
The federal government’s abject failure to create any form of meaningful AI oversight or accountability should alarm every American
Newsom is asking Congress and President Trump to review California's framework and adopt it, or to set the state's rules as the minimum for the rest of the country. The release puts that beside Trump's rejection of new AI regulations.
Granted, nothing in the order requires a company to build a kill switch today. It requires a recommendation about one, two months from the day he signed it.