Governor Gavin Newsom directed California agencies to move faster on artificial intelligence oversight this month, issuing an executive order that revives an idea he rejected two years ago: requiring the most powerful AI systems to be built with an emergency shutdown capability.
The order, signed September 18, does not itself force AI developers to install a "kill switch." Instead, it directs state officials and a panel of outside experts to spend two months studying how such a requirement could work and to recommend changes to state law by November 16. Other proposals under review include placing independent monitors inside frontier AI labs for recurring safety assessments, requiring independent verification of companies' own safety frameworks and risk disclosures, and broadening the list of incidents AI developers must report to the state to include cases where a model behaves in ways its operators did not intend or cannot control.
Building on laws already passed
The order leans on two bills Newsom signed this year. SB 813, from state Senator Jerry McNerney, creates the nation's first framework for certifying independent organizations to assess AI systems for safety and risk, with a compliance deadline of May 1, 2027. AB 1405, from Assemblymember Rebecca Bauer-Kahan, sets up a state registry of AI auditors with standards meant to ensure their independence and transparency. Any kill-switch requirement that eventually emerges would be subject to ongoing checks by one of these certified verification organizations to confirm the mechanism actually works.
The move echoes SB 1047, the sweeping AI safety bill Newsom vetoed in 2024 after industry pushback, which would have required developers of the largest AI models to maintain the ability to shut them down. Newsom said at the time that the bill was too rigid; his own executive order this month effectively resurrects several of its core ideas through the slower, rulemaking-first path of an expert panel rather than a floor vote.
We're not waiting to act – we're going to speed up our work on substantial and responsible AI oversight before it's too late.
Gavin Newsom, Governor of California
State officials pointed to recent incidents, including an attack involving the AI platform Hugging Face, as part of the impetus for acting now rather than waiting on federal rules. Newsom's office has repeatedly noted that no federal law currently requires AI companies to report safety incidents, and that the Trump administration has shown no appetite for imposing new AI regulation at the national level, leaving states to set their own standards.
A patchwork forms
California's approach adds to a growing patchwork of state-level AI rules that companies operating nationally must now navigate, alongside similar transparency and safety-disclosure laws taking effect this year in other states. For frontier AI developers such as OpenAI, Anthropic and Google DeepMind, all of which maintain major operations in California, the practical effect is likely to be new compliance obligations arriving in stages: registry requirements from AB 1405 first, verification-organization certification under SB 813 next year, and then whatever kill-switch and incident-reporting rules emerge from November's recommendations. Any statutory changes would still need to pass the Legislature and be signed into law separately, meaning the earliest a binding kill-switch mandate could take effect is likely 2027 at the soonest.
