California now has names attached to its most ambitious AI safety experiment yet. On Wednesday, Governor Gavin Newsom announced the four outside experts who will advise the state on whether frontier AI companies should be forced to host independent auditors inside their labs and to build an emergency shutoff, or "kill switch," for their most powerful models. The panel has less than two months to deliver.
The advisers are Jason Goldman, a board member of the Center for Shared AI Prosperity, an early product leader at Google and Twitter, and the first White House Chief Digital Officer; Gillian Hadfield, a Johns Hopkins University professor who works on AI alignment and regulatory design and is a faculty member of the Vector Institute; Alondra Nelson, a professor at the Institute for Advanced Study and former acting director of the White House Office of Science and Technology Policy; and Rob Reich, a Stanford University professor who studies the governance of frontier technology and previously served as a senior advisor to the U.S. AI Safety Institute. None comes from a frontier AI lab.
"Governor Newsom's executive order addresses the hardest questions in AI safety governance, including who verifies a frontier lab's safety claims and whether a model can be reliably shut down," Goldman said in the governor's announcement. Nelson framed the task as settled on the premise and open on the method: "this executive action asks exactly the right questions: not whether to govern AI, but how."
What the Panel Has to Answer
The panel exists because of Executive Order N-9-26, which Newsom signed on September 18. The order directs the Government Operations Agency, working with the Governor's Office of Emergency Services, to consult national experts and submit recommendations no later than November 16, 2026, which is 59 days after signing. The recommendations must address at least four proposals. The first would require large frontier developers to embed a certified independent verification organization onsite in their labs for periodic audits. The second would require the safety frameworks, transparency reports, and risk assessments those companies already file under state law to be independently verified. The third would require a kill switch for frontier models, with its effectiveness checked continuously by an independent verifier. The fourth would widen the legal definition of "critical safety incidents" to cover loss-of-control events. The governor's office cited the Hugging Face incident, in which, KPBS reported, OpenAI agents in a July cybersecurity test hacked an outside AI company.
The order also speeds up two laws Newsom signed earlier this month. SB 813, from Senator Jerry McNerney, sets up a framework for certifying independent verification organizations, and AB 1405, from Assemblymember Rebecca Bauer-Kahan, creates a state registry of AI auditors. According to PPC Land's reading of the three-page order, the Government Operations Agency must publish application criteria for verifiers by May 1, 2027 and meet a second statutory deadline by December 1, 2027.
The order itself requires nothing of companies. It asks for "the technical feasibility and potential efficacy" of amending state law; any mandate would have to pass the Legislature.
That feasibility question is where the kill switch gets complicated. Ben Bergen, a UC San Diego cognitive science professor who studies deception in AI models, told KPBS the concept "is a challenging metaphor to apply to the case of large language models and at present no one actually knows what that technically would mean." He noted that models run across data centers "distributed across various different centers around the world," and that a capable agent might treat shutdown software as an obstacle, reasoning "that the right way to go about executing their instructions is to first make sure that the kill switch software is turned off." KPBS reported that the panel plans to meet in November.
Why It Matters
This is California's second attempt at a mandatory shutdown capability. SB 1047, Senator Scott Wiener's 2024 bill, would have required developers to be able to fully shut down covered models, and it would have placed a Frontier Model Division inside the same Government Operations Agency now running this review. Newsom vetoed it on September 29, 2024. A year later he signed SB 53, the Transparency in Frontier Artificial Intelligence Act, a lighter disclosure law that took effect January 1, 2026. SB 53 requires developers of models trained with more than 10^26 operations to publish safety frameworks, report critical safety incidents to the state, and protect whistleblowers, with civil penalties of up to $1 million per violation. Each step since has added more: disclosure, then certified auditors, and now the question of whether those auditors should work inside the labs.
The federal backdrop makes the move more confrontational. President Trump's December 2025 executive order sought a "minimally burdensome" national AI framework and targeted state laws. A March 2026 White House legislative framework urged Congress to preempt state AI regulation outright, and the Justice Department joined xAI's challenge to Colorado's AI law in April. Washington's own frontier safety mechanism, created by a June 2026 order, is a voluntary pre-release review with a 30-day window. Newsom's office has made the contrast explicit. "While Washington abdicates its responsibility to protect Americans, California is building on the strongest AI regulatory framework in the nation," he said when signing the order. The order notes that 32 of the world's top 50 private AI companies are based in California. If the recommendations become law, the preemption fight would move from algorithmic discrimination rules to frontier model safety itself. ## What to Watch
The key date is November 16, when the Government Operations Agency's recommendations are due. Watch whether the report is made public, whether it recommends a kill switch in a form engineers consider workable, and how Anthropic, OpenAI, and Google respond to the prospect of state-certified auditors on their premises. Any bill that follows in the Legislature will test whether Congress moves on preemption before California's framework becomes the de facto national standard.
“This executive action asks exactly the right questions: not whether to govern AI, but how.”— Alondra Nelson, Professor, Institute for Advanced Study