Newsom Signs Executive Order Pushing AI Safety Measures, Including a Possible Kill Switch

California Gov. Gavin Newsom signed an executive order on Friday, September 18, 2026, that he said would sharply speed up AI safety work in the state, The Washington Post reported. The order could lead to a requirement that companies build an emergency shutoff into advanced AI models.

Newsom argued that federal regulators have not done enough, according to the Post. That framing puts the nation’s most populous state, and the home of many of the largest AI companies, in the position of moving ahead of Washington on AI safety.

A kill switch, again

The possible shutoff requirement echoes a debate in the US Senate two days earlier. On September 16, Sen. John Kennedy (R-La.) tried to pass his AI Emergency Button Act by unanimous consent, and Sen. Rand Paul objected; see our report on the Senate kill-switch bill. Kennedy’s bill would have required developers selling advanced AI models in the US to build in an emergency shutoff controlled by the companies themselves.

California’s order is a different instrument. An executive order directs the state’s own agencies; it is not a statute passed by the legislature. Based on the Post’s report, the order could lead to a kill-switch requirement rather than imposing one immediately, which means the details of any future rule still have to be written.

Why California matters

Because so many AI developers are based in California, a state-level requirement there could shape practice across the industry even without federal action. Newsom’s argument that federal regulators have fallen short also signals that the tension between state and federal approaches to AI is likely to keep growing.

That tension became sharper in the following days. President Trump announced plans for an “AI Force” and an AI czar while saying he would not hinder the industry’s growth; see our report on the AI Force announcement. California’s attorney general later served OpenAI with an investigative subpoena over cybersecurity incidents involving its models; see our report on the subpoena.

What the source does not yet settle is just as important. The report summarized here does not list the specific deadlines in the order, which models would count as “advanced,” or who would decide when a shutoff must be used. Until the state publishes those details, the order is best read as a statement of direction from the governor’s office rather than a finished rulebook that companies must already follow.

What to watch

The next steps to follow are which state agencies the order directs, what timelines they are given, how an emergency-shutoff requirement would be defined and enforced, and whether AI companies or federal officials push back on a state-level approach.