AI Kill-Switch Folly

California may police what happens, but may not write the operating system for a product that exists for the entire country

California Governor Gavin Newsom’s latest folly, ordering a kill switch on AI, does not require OpenAI, Google, Anthropic, or anyone else to build a working emergency off button.

Click Here to Subscribe to the California Political Review

It instead tells California agencies to hurry two bills that were already law – and to hire a two-month expert panel to consider kill switches. It also broadens the definition of when an AI-related incident happens, requiring companies yell “oops.”

That is not a highly-publicized safeguard. That is a study group with better lighting and a noun that polls well.

The legal problem: California may police what happens, well, to Californians. It may not write the operating system for a product that exists for the entire country. AI technology is not executed differently in Sacramento and Omaha.

If Newsom’s order ever becomes a statute – forcing labs to redesign or shutdown architecture for California users, it walks into the Commerce Clause of the U.S. Constitution wearing a neon sandwich board.

Any apparently neutral state law still dies when the burden on interstate commerce is “clearly excessive” relative to the local benefit. Newsom’s “Put our verified emergency off switch in the national model” is the proof.

The White House has put up a Justice Department task force to sue states over exactly this theory.

Sacramento cannot preempt federal power with a press release like Newsom’s.

California cannot command a processing cluster in Texas, a lab in London, or files sitting on a server that has never heard of the Government Operations Agency.

But it can squeeze firms that want California customers or already live in the Bay Area. That is a shakedown, not sovereignty.

The obvious reply from any AI lab that does not want state government monitors embedded on site is: we will train somewhere else, thin out the California interface, and treat Newsom’s rule as prohibitive.

Foreign developers make Newsom’s grand decree even thinner. A Chinese, Gulf, or European champion that never serves the California is simply not in his statute. Newsom’s order does not fix this.

California cannot place “independent verifiers” in Shenzhen or make Beijing maintain a California-certified emergency off switch. AI labs that treat the California market as optional will simply not RSVP.

After Newsom’s press release, it’s obvious: a kill switch that only reaches companies still willing to sit in the room is not a kill switch for “AI.” It is a kill switch for the people serving California who showed up.

Even on its own terms, the plan is built to miss. The evidence Newsom is waving- the July 2026 AI containment failure- did not trigger the statute he already signed. Officials said it missed SB 53’s reporting threshold.

A million-dollar civil penalty per incident is not “steep” to an AI lab; it’s an expense line item next to snacks.

And a “verified” kill switch sounds tough until someone asks what happens to an AI program copied, fine-tuned, serving California from six outside regions. If the suspicious AI capability is already out there, shutting off the original developer’s program becomes whack-a-mole.
Click here to read the full article in the California Globe