Gavin Newsom demands AI kill switch in California
California Governor Gavin Newsom issued an executive order to establish strict AI safety rules, including a potential mandatory kill switch, stepping in where federal efforts have stalled.
Governor Gavin Newsom signed an executive order on Friday directing California to assemble a group of experts to recommend new AI safety measures within two months. The order aims to establish a robust regulatory framework for frontier models, positioning California to lead on tech oversight while federal legislation remains stalled in Congress.
Under the directive, the expert panel will evaluate several stringent safety mandates. These include requiring AI developers to host onsite independent verification teams for routine audits and subjecting risk assessments to independent auditing standards. Newsom also wants the state to explore mandating a routinely verified kill switch for AI systems and requiring companies to report loss-of-control incidents—such as the OpenAI attack on Hugging Face—as critical safety events. Additionally, the order fast-tracks the implementation of two recently signed state laws that establish a registry of AI auditors and a framework for independent safety verifiers.
The push comes amid heightened anxiety over AI safety, highlighted by former Anthropic researcher Jacob Coxon warning of a greater than 10 percent chance that AI could destroy humanity within a decade. With the U.S. House of Representatives on recess until November and midterm elections approaching, federal action is highly unlikely. Meanwhile, Donald Trump has dismissed the need for legislative guardrails, claiming a strong president is the only necessary control. Newsom criticized Washington's inaction, calling it an "abject failure" and urging federal leaders to adopt California's rules as a baseline.
For AI practitioners and developers, California's aggressive stance signals an imminent shift toward heavy compliance burdens. If these recommendations become law, companies building frontier models will need to prepare for constant external surveillance, formal risk reporting, and the technical integration of fail-safe shutdown mechanisms. Because California is the hub of the U.S. technology sector, these local mandates will likely dictate the operational standards for the entire global AI industry.
This is our own summary of reporting by The Verge AI
