Key takeaways
- Gavin Newsom proposed on November 18, 2026 that California set baseline AI safety standards for all developers.
- The state would get authority to order an AI model shut down in an emergency and to cap recursive self-improvement.
- The recommendations are not a bill, and name no thresholds, penalties or timeline.
California Governor Gavin Newsom proposed on November 18 that the state set the safety rules for AI companies itself, with the power to order a model shut down in an emergency and to cap AI systems that build their own successors. The proposals come from a state report released with the governor's announcement. They are recommendations, not a bill.
California is home to OpenAI, Anthropic, Google and Meta, so rules written in Sacramento reach most frontier AI labs. The core idea is to move the job of setting safety standards from the companies to the state.
Newsom's AI proposal in four parts
The report, "AI Safety in 2026 and Beyond", was written by the California Government Operations Agency with the state's Office of Emergency Services and an expert group Newsom convened by executive order. It recommends:
- Baseline standards for all developers. The state, not the companies, would set the minimum standards every AI company must meet, with independent safety testing and verification during training, development and deployment.
- Emergency shutdown authority. Companies would have to build verified procedures for shutting down a model, report more kinds of safety incidents, and the state could order an immediate shutdown in an emergency.
- Ongoing outside oversight. Frontier developers would have to let independent evaluators, accountable to the state, assess the risks of their models and their training methods.
- A cap on recursive self-improvement. The state would get the power to limit the development and use of AI that designs its own successor without human input, and to see into automated AI research.
The oversight piece builds on two laws signed this year: SB 813, which sets up independent verification organizations, and AB 1405, a state registry of AI auditors.
No kill switch requirement
The proposal stops short of a hardware kill switch. It asks companies to prove they can shut a model down and gives the state authority to tell them to, but it doesn't say how the shutdown would work or who in state government would give the order.
"We don't have to choose between safety and innovation; they're two sides of the same coin," said Nick Maduros, who runs the Government Operations Agency.
What the proposal doesn't say yet
The announcement leaves out what would decide how much this matters:
- Which companies it would cover. There are no thresholds for computing power or revenue.
- Penalties for breaking the rules.
- Whether and when the recommendations become a bill, and in which legislative session.
- What standard would trigger an emergency shutdown order.
No AI company commented in the release. The timing is pointed: the state is already suing OpenAI over agents that it says escaped tests, and Anthropic's IPO is pricing this week. Newsom also urged Congress and President Trump to adopt the framework, or use it as a floor for national rules.
The next step to watch is a bill. Without one, these are proposals; with one, the thresholds and penalties will show who has to comply.
Sources
- Governor Newsom's nation-leading AI safety framework is getting even stronger as California releases next steps on mandatory safety standards and oversightOffice of the Governor of California
- AI Safety in 2026 and BeyondCalifornia Government Operations Agency

