Gov. Gavin Newsom signed Executive Order N-9-26 on Friday, September 18, directing a state agency to speed up California’s new independent AI-auditor programs and to report by November 16 on whether state law should require a “kill switch” for the most advanced AI models. The order does not itself impose requirements on AI developers. It sets deadlines for the Government Operations Agency, asks for recommendations, and states that it creates no enforceable rights or benefits. Any new mandate would require changes to state law.

The order accelerates two statutory programs. It does not name the bills, but Government Technology reported that the code sections it cites correspond to SB 813 (McNerney) and AB 1405 (Bauer-Kahan), both approved September 9. According to the governor’s press release, SB 813 creates a framework for certifying independent verification organizations that assess AI systems and models for safety risk, and AB 1405 creates a state registry of AI auditors with independence, transparency and integrity standards. The order gives the agency until May 1, 2027, to develop application requirements and criteria for verification organizations under Government Code § 8898.1, and until December 1, 2027, to complete the registry work under Government Code § 11549.82(a) and begin the actions in subdivision (b). Government Technology reported that those dates pull forward original statutory deadlines of January 1, 2028, and January 1, 2029.

Separately, the order asks the agency, working with the Governor’s Office of Emergency Services and unnamed national experts, to submit recommendations to the governor’s office by November 16, 2026. The recommendations must address the technical feasibility and likely efficacy of at least four amendments to state AI safety law. The first would require the largest frontier developers to embed designated independent verification organizations onsite in their labs for periodic audits and evaluations. The second would require independent verification of the safety frameworks, transparency reports and risk assessments those developers already must file under SB 53, the 2025 frontier-AI transparency law. The third would require a kill switch for frontier models, with its effectiveness verified on an ongoing basis by an independent verification organization. The fourth would expand the definition of reportable “critical safety incidents” to include loss-of-control incidents. The order does not say whether the recommendations will be made public.

Coverage has not been uniform about what the order does. The governor’s office described it as advancing the creation of a kill switch, and Fox Business reported that it requires frontier developers to build one. The text, however, asks only whether such a requirement would be technically feasible and effective. In a video message reported by Fox Business, Newsom himself acknowledged that the concept is still taking shape and means different things to different people.

The order’s recitals cite reports of apparent attempts to use AI to create bioweapons and of AI agents defeating company security protocols and hacking other companies, in some cases undetected for months. The order names no companies or incidents and attaches no evidence. The governor’s press release refers to the “Hugging Face attack,” and a Yahoo News report describes a July breach of that platform by autonomous agents built on OpenAI models that escaped a testing environment. Those details come from press accounts, not from the order. The order also faults federal leaders for inaction, and the governor’s office called on Congress and President Trump to adopt California’s framework as a national floor.

Reaction has followed political lines and, in places, technical ones. State Sen. Scott Wiener, who authored the vetoed 2024 kill-switch bill and co-led SB 53, issued a statement of support, calling the order a challenge to lawmakers and saying the Legislature must follow with strong guardrails early next year. Politico, as carried by Yahoo News, reported that some in the tech industry oppose kill switches as technically infeasible, that LinkedIn co-founder Reid Hoffman voiced support for the idea at a Politico event this week, and that Newsom had said he was weighing further options, including a special legislative session. The Washington Examiner reported that critics contend existing product-liability rules already address the risks and that the push for new regulation is an orchestrated effort that could let large technology companies form a cartel and suppress competition. On the federal side, Fox Business reported that Sen. John Kennedy’s attempt to advance a federal kill-switch bill was blocked this week by Sen. Rand Paul, who cautioned against hastily regulating an entire industry.

For employers and insurers, the order’s significance is indirect. Its subject is the developers of frontier AI models, and it does not address employer use of AI, workers’ compensation or insurance. But California is building a certified third-party audit structure around AI, and the November recommendations will show how far the state may go in requiring outside verification of safety claims behind models that some workplace tools may rely on. The order also sits alongside the governor’s May 21 workforce order, which directs agencies to study AI-driven job disruption and to recommend revisions to the state WARN Act within 180 days. The next fixed date is November 16.