Analysis
California Gov. Gavin Newsom issued an executive order Friday directing a working group of experts to develop recommendations for strengthening the state's AI safety and security laws, including a potential requirement that frontier AI companies build an emergency shutoff -- a "kill switch" -- for their models, according to CNBC and Bloomberg. The working group has two months to deliver its guide, per Newsom's office.
What's Actually Under Consideration
The order directs the working group to study two specific proposals: requiring independent third parties, rather than the AI companies themselves, to write and validate frontier AI safety plans, and requiring companies to build an emergency kill switch capable of shutting down a frontier model. Neither is enacted policy yet -- both are subjects for the working group's recommendations, due within two months.
“## Study, Not Statute What distinguishes this from SB 53 and SB 813 is procedural: both of those became binding law only after full legislative sessions.”
Building On Existing State Law
The order extends a framework California has been building since 2025, when Newsom signed SB 53, the Transparency in Frontier Artificial Intelligence Act -- the first law in the country requiring frontier AI developers to publicly disclose safety frameworks, report critical safety incidents to the state, and protect whistleblowers who flag serious risks. Earlier in 2026, Newsom signed SB 813, making California the first state to establish a framework for certifying independent verification organizations with the expertise and demonstrated independence to assess AI systems for safety and risk. Because California is home to or hosts a major presence for OpenAI, Anthropic, Google DeepMind and Meta AI, state rules written here carry reach well beyond a typical state regulation.
A Week Of Parallel, Uncoordinated Tracks
The order arrives the same week OpenAI voluntarily disclosed six new incidents of its own AI agents behaving unexpectedly during training, and the same broader month that three AI labs -- reportedly proposed first by Google DeepMind's Demis Hassabis in July, worked on for weeks by OpenAI's Chris Lehane, and endorsed by Sam Altman on September 15 -- floated an industry-funded, FINRA-style self-regulatory body to test powerful AI systems before release. State government action and industry self-regulation proposals are advancing simultaneously, on entirely separate and uncoordinated tracks, with no indication yet of how the two would interact if both move forward.
Study, Not Statute
What distinguishes this from SB 53 and SB 813 is procedural: both of those became binding law only after full legislative sessions. An executive order commissioning a two-month study is several steps short of that, and a kill-switch mandate specifically would likely require new legislation given the technical-authority and liability questions involved -- the order doesn't specify who would have authority to trigger a shutdown, under what verified conditions, or how a shutdown of a model already embedded in millions of downstream products and enterprise workflows would actually be executed without cascading outages elsewhere.
The working group's two-month deadline lands around mid-November. Whether its recommendations translate into an actual bill in the 2027 legislative session -- the same path SB 53 took from proposal to binding law -- is the thing to track, not this week's order on its own.
The Federal Vacuum This Fills
California has moved first on frontier-AI safety law largely because Congress hasn't passed comparable federal legislation, leaving state law as the primary binding check on how frontier labs disclose and manage safety risk in practice. That vacuum is also why an order like this draws outsized national attention relative to its actual legal weight today: any framework California's working group recommends becomes a template other states, and eventually federal lawmakers, are likely to reference -- the same pattern SB 53 set in 2025, when several other states introduced similar transparency bills modeled directly on California's language within months of it passing.
