California Wants an Off Switch for the Most Powerful AI
California Gov. Gavin Newsom issued an executive order on Friday, September 18, 2026, to put California in the lead on AI oversight and start work on a possible off switch for the most advanced AI systems. The Verge
The order tells the state to bring together experts. They must send back ideas within two months for stronger AI safety laws.
The job is narrow. It is about rules that can be enforced by law, not voluntary promises.
What the order puts on the table
The experts will look at placing outside checking teams inside AI companies for regular reviews.
They will also look at making company safety reports follow standards set by outside reviewers.
A third idea is the off switch itself. Like a circuit breaker for a house, it would let developers shut down certain AI programs in an emergency. The switch would have to be tested on a regular basis to show it works. As described in state coverage, developers would be required to turn off certain programs in emergencies. CalMatters
A fourth idea would have outside specialists write the safety plans for companies that build the most advanced AI, according to the Governor's Office release. Governor's Office
The order also tells a state agency to move faster on two recent laws. One sets up a system for outside checkers to review AI safety. The other creates a state list of approved AI reviewers.
The Governor's Office published the action on September 18, 2026, under the title "Governor Newsom issues executive order to accelerate independent oversight and advance the creation of an AI kill switch."
Newsom asked Congress and President Donald Trump to study California's plan and use it as a model for federal action.
Two days before the order, on Wednesday, Newsom said in an interview with Politico that he might call state lawmakers back for a special session on AI.
The broader context here is a run of state steps. On September 9, 2026, Newsom signed SB 813, described as first-in-the-nation AI safeguards to protect Californians. A separate California order says firms seeking state contracts must have safeguards against abuse. Reuters In September 2024, Newsom vetoed a disputed AI safety bill after objections from the tech industry.
Why verification will decide this
The key word in the order is independent. Outside teams on site, outside rules for reports, and safety plans written by outsiders all point to the same choice.
Safety would no longer rest on internal tests and company-written summaries. It would depend on outside parties with set access and the power to test the off switch.
In my view, that outside checking matters more than the phrase kill switch. A shutdown rule is simple to say and hard to carry out.
Copies of a model can spread. Adapted versions can live on other computers. AI helpers that use tools and remember past steps are hard to stop at once. The meaning of "shut off certain programs" will decide if this means stopping new uses, halting training, or pulling the service offline.
The broader context here is that the two-month review must settle plain definitions first. The state still needs rules for who can serve as a reviewer, what test tools and incident records they can see, how often rechecks happen, and who is liable if shutdown fails.
In my experience watching safety rules spread, buyers will feel this first in purchasing. California already ties state contracts to safeguards. If outside grades become normal, they will show up in vendor checklists and contracts for shutdown and incident response. Similar security checklists such as SOC 2, FedRAMP, and zero-trust started narrow and later became standard.
In my view, after covering shifts from client-server to cloud to mobile to AI, this can work if it stays practical. Outside review helps when reviewers can look inside, repeat tests, and interrupt service. I watched my kids grow up with each new platform, and clear rules helped schools and parents adapt faster. Clear and testable shutdown rules would let builders move faster under clearer guardrails.
