California Governor Gavin Newsom ordered state agencies Friday to accelerate a new independent AI-oversight system and study whether frontier artificial-intelligence models should be required to include an independently verified emergency shutoff.
The action does not itself create an AI “kill switch” or immediately impose new duties on developers. Instead, Executive Order N-9-26 directs the Government Operations Agency, working with the Governor’s Office of Emergency Services and national experts, to recommend possible changes to state law by November 16. Those recommendations must address the technical feasibility and likely effectiveness of several oversight proposals.
The required review includes placing designated independent verification organizations inside large frontier-model laboratories for periodic audits; independently checking company safety frameworks, transparency reports and risk assessments; testing an emergency shutoff on an ongoing basis; and broadening reportable critical incidents to include loss-of-control events. The governor’s announcement characterizes these ideas as proposals under consideration, an important distinction from enacted standards.
The order also sets implementation deadlines for two laws Newsom signed earlier this month. It tells the Government Operations Agency to publish application requirements, procedures and criteria for independent verification organizations by May 1, 2027, and to begin specified regulation of AI auditors by December 1, 2027. A California Board of Accountancy analysis of AB 1405 described the auditor framework as establishing enrollment, transparency, competency and ethical standards while also flagging potential overlap, inconsistency and implementation costs.
SB 813 creates the separate framework for independent verification organizations. During the Assembly’s final consideration, supporters said those organizations must have relevant expertise, credible methods and independence from the companies they assess, according to the official floor-hearing transcript. The order now pushes the administration to translate that framework into operational rules on a faster schedule.
The policy arrives amid a broader argument over whether voluntary company commitments are enough. Earlier in September, OpenAI called for mandatory national requirements covering testing, independent assessment, cybersecurity and incident reporting for the most advanced systems, while backing California’s new verification and auditor laws. That industry support does not resolve how an emergency shutoff would work, who could activate it or how regulators would test it without creating a new vulnerability.
The executive order places that technical debate inside California’s existing regulatory structure. It notes that 32 of the world’s 50 leading private AI companies are based in the state, framing the review as both a public-safety measure and an economic-policy decision. The document also says the recommendations should focus on amendments to existing law, rather than treating the study as a freestanding enforcement program.
The near-term consequence is therefore procedural but significant: California has established firm agency deadlines and a two-month window for a technical regulatory blueprint. Any binding kill-switch mandate, onsite-verifier requirement or expanded incident-reporting rule would still require additional legal action after the recommendations are delivered.
