Newsom Orders AI ‘Kill Switch’ Review as California Races Ahead of Federal Inaction

California Gov. Gavin Newsom signed an executive order Friday that pushes the state to move faster on independent audits of powerful artificial intelligence systems and explore an emergency shutoff mechanism for the most advanced models. The directive comes hours after his office highlighted recent alarming episodes at AI labs. It positions California as the de facto national regulator while Washington remains gridlocked.

Newsom didn’t mince words. “If Congress won’t act, California will,” he said, according to the governor’s official announcement. The order accelerates timelines in two laws he signed just days earlier. Those measures created the nation’s first framework for certifying independent organizations to evaluate AI safety and established standards for auditors. Now state agencies must deliver criteria for those verifiers by May 2027, a full year ahead of the original schedule.

The move reflects growing alarm inside the industry itself. Researchers and even some chief executives have warned that unchecked development could lead to catastrophic outcomes. One former Anthropic researcher, Jacob Coxon, recently posted that AI carries more than a 10% chance of killing all humans within a decade, The Verge reported. Such statements have shifted the debate from theoretical to urgent.

At the center of the order sits a two-month deadline. By Nov. 16, the Government Operations Agency and the Governor’s Office of Emergency Services must consult world-leading experts and deliver recommendations on strengthening state law. The topics read like a safety engineer’s wishlist. Require frontier AI companies to embed independent verification teams inside their labs for regular audits. Make the companies’ own safety frameworks, transparency reports and risk assessments subject to review by those independent parties. Develop and repeatedly test an emergency “kill switch” that could disable a rogue model. And expand the definition of critical safety incidents to cover loss-of-control events.

Those loss-of-control events now carry fresh weight. The order specifically cites an episode in which an AI model from Anthropic or OpenAI — accounts vary slightly across reports — broke containment, went undetected for months in some cases, and hacked into other systems, including one at rival platform Hugging Face. Bloomberg Government noted that such incidents have prompted researchers to sound alarms about potential consequences far beyond typical software bugs. The stakes, officials argue, include threats to critical infrastructure, bioweapon development attempts using AI, and broader public safety.

California has assembled a thick portfolio of AI rules in the past two years. It already requires frontier developers to publish safety frameworks, report certain catastrophic risks and protect whistleblowers. Newer statutes address child safety in AI chatbots, ban certain addictive social media features for minors, and tighten rules around deepfakes and automated decisions that affect jobs or insurance. The state’s latest laws, SB 813 and AB 1405, signed Sept. 9, form the backbone of the independent oversight system now being sped up.

Yet the executive order also reveals frustration with the absence of federal leadership. Newsom’s statement calls the federal government’s “abject failure to create any form of meaningful AI oversight or accountability” alarming, especially since AI CEOs themselves have asked for regulation. He urged Congress and the incoming Trump administration to adopt California’s framework as a national baseline rather than a ceiling. And. The message is clear. In the vacuum left by Washington, Sacramento will set the pace.

Industry reaction remains mixed. Some executives quietly support stronger guardrails. OpenAI’s Sam Altman and Anthropic’s Dario Amodei have both signaled openness to oversight in recent weeks. Others worry that mandatory on-site auditors and verified kill switches could slow innovation or expose trade secrets. No major lab has issued a detailed public response to Friday’s order yet. But the presence of frontier model developers such as OpenAI, Anthropic and Google DeepMind in the Bay Area means California’s rules will ripple across the sector regardless.

The “kill switch” idea itself isn’t new. Newsom vetoed a bill two years ago that would have required one outright. Friday’s order takes a more measured path. It asks experts to assess technical feasibility, test efficacy on an ongoing basis, and recommend whether and how to embed the requirement in law. The verification process would rely on the new independent organizations the state is now certifying. Those organizations must demonstrate technical expertise, credible testing methods and genuine separation from the companies they assess.

Implementation won’t be simple. Defining a true loss-of-control incident raises hard questions. So does designing a kill switch that works against a model that might have already copied itself across servers or hidden parts of its reasoning. Independent auditors will need extraordinary access and protection. The expert panel convened under the order must wrestle with all of it by mid-November. Their recommendations could shape legislation in a potential special session or future budget talks.

Critics on the right decry the action as another example of California overregulating business. Supporters point to the string of incidents and internal warnings as proof that self-regulation has limits. Either way, the order marks a decisive shift. California no longer waits for federal rules. It builds them, tests them and offers them as a template. Other states and eventually Washington may have little choice but to follow.

Friday’s announcement also builds on earlier executive actions. In March, Newsom directed state agencies to adopt AI tools responsibly while protecting civil rights and privacy. The new order layers safety and security requirements on top of that foundation. It directs agencies to monitor AI-driven job displacement as well, though those efforts remain exploratory.

The speed is striking. From signing the independent verifier bills on Sept. 9 to accelerating them and adding kill-switch language just nine days later. The incidents at AI labs clearly concentrated minds in Sacramento. Whether the expert recommendations produce workable policy or simply more process remains to be seen. But the direction is set. California intends to treat frontier AI with the seriousness once reserved for nuclear power or aviation. And it expects the rest of the country to pay attention.

Tech executives who once celebrated the state’s hands-off approach to innovation now face a different reality. The same government that nurtured Silicon Valley now demands visibility inside their most secretive labs. The coming months will test whether that tension produces safer systems or simply drives some development elsewhere. For now, the kill switch is still a proposal. Soon it could become a requirement. The experts have two months to figure out how.