Newsom Orders California to Draft AI 'Kill Switch' Rules Within Two Months

California Gov. Gavin Newsom signed an executive order Friday directing experts to recommend how the state could force frontier AI models to shut down and report safety incidents.
California Gov. Gavin Newsom signed an executive order Friday that puts the state at the center of AI oversight, including a possible requirement that developers of frontier models build in a mechanism to shut them down. The order directs state agencies to assemble a panel of outside experts and deliver recommendations within two months on how California should regulate the most powerful AI systems. It marks one of the most aggressive moves by a US state to claim authority over a technology that has so far been governed mostly by voluntary corporate pledges and federal guidance.
The centerpiece is what the order calls a "kill switch" — a capability that would let authorities or developers halt a model that is behaving dangerously. The expert group is also expected to address reporting requirements for safety incidents, so that failures or misuse are disclosed rather than handled privately. Newsom's office framed the directive as a way to move faster than Washington, which has yet to pass comprehensive AI legislation. California is home to OpenAI, Anthropic, Google and Meta, giving the state unusual leverage over the companies building the largest models.
The two-month timeline is unusually tight for a process that typically takes a year or more. It signals Newsom wants concrete proposals before the legislature reconvenes in earnest and before the 2026 election cycle consumes the agenda. The order does not itself impose any rule; it sets up a framework that could become legislation or emergency regulation. That distinction matters, because a mandate issued through the governor's office would face immediate legal challenges from industry groups that argue only the federal government can regulate interstate software.
Industry reaction is likely to split along familiar lines. Smaller AI safety organizations have pushed for exactly this kind of binding authority, arguing that voluntary commitments have not prevented models from being released with known risks. Larger developers have warned that a patchwork of state rules would slow releases and push work to other jurisdictions. The order's expert panel will have to weigh those arguments while producing something specific enough to survive a court fight.
California already regulates consumer privacy and online safety in ways that set national norms, and companies often comply statewide rather than build separate products. If the recommendations turn into a kill-switch requirement, the practical effect could reach well beyond the state's borders. The open question is whether two months of expert testimony can produce a technically workable standard — and whether Newsom will push it through the legislature or try to enact it on his own authority. Watch for the panel's membership, which will indicate how seriously the state intends to challenge the major labs.