California governor pushes kill switch for AI models

Governor Gavin Newsom signed an executive order Friday that places California at the forefront of U.S. AI oversight, including the potential to mandate a "kill switch" for advanced models. The move comes while Congress is gridlocked, the House is in recess until November, and the midterm elections make significant Senate legislation unlikely in the near term.

The order convenes an expert group to deliver recommendations within two months for strengthening safety measures in state legislation. Items on the agenda: requiring companies to embed independent verification teams on site for ongoing audits; subjecting transparency reports and risk assessments to external-auditor standards; creating a regularly tested shutdown mechanism; and mandating disclosure of "loss-of-control events" — such as the Hugging Face breach attributed to OpenAI — as critical safety incidents.

At the same time, the order accelerates implementation of two laws Newsom has already signed: a framework for independent auditors to evaluate model safety, and a state registry of AI auditors. Both are designed to build institutional infrastructure for continuous, not one-off, oversight.

Newsom frames the order as a federal precedent and calls on Congress and President Donald Trump to "examine and adopt the state's framework, or use our leading regulation as a floor, not a ceiling." In a statement he argued that "the total failure of the federal government to create any form of meaningful oversight should concern every American, especially when AI CEOs themselves are begging for regulation."

In a Politico interview Wednesday, Newsom signaled he may convene a special legislative session on the issue, alongside further executive actions. The backdrop is a warning from former Anthropic researcher Jacob Coxon, who put the probability at more than 10% that AI would "kill all humans" within a decade — a claim that ignited congressional debate but produced no legislation.