California asks experts to assess AI safety rules and a possible kill switch

The governor announced experts to advise California’s work on AI, and the order asks them to assess potential changes to existing state AI safety and security laws.[1][2] Among the ideas under consideration is requiring a “kill switch” for frontier models, with its effectiveness independently verified on an ongoing basis.[2] Axios reports that officials and experts are being asked to recommend how such a requirement could work; the order does not immediately install a shutdown system.[4] The recommendations, developed in consultation with national experts, are due no later than November 16, 2026.[2]

The order calls for considering whether large frontier AI developers should embed independent verification organizations in their labs for periodic audits and evaluations.[2] It also calls for independent verification of safety frameworks, transparency reports and risk assessments that frontier AI companies are required to file.[2] Officials are asked to consider expanding reportable critical safety incidents to include a range of loss-of-control events.[2] The recommendations are meant to address both the technical feasibility and potential effectiveness of changes to existing state laws.[2]

The order accelerates implementation timelines for California’s SB 813 and AB 1405.[3] Separately, California’s 2025 SB 53 requires frontier AI developers to disclose safety frameworks, report specified critical safety incidents to the state and protect whistleblowers who report serious risks.[3] The new recommendations would examine possible changes to existing laws, rather than establish that the proposals have already been adopted.[2][4] A previously vetoed measure would have required the largest AI developers to undergo external safety auditing and develop a kill switch.[6]

Newsom said California was bringing experts together to build on the state’s AI framework and address public concerns.[1] Jason Goldman, identified as a board member of the Center for Shared AI Prosperity, said the order raises questions about who verifies frontier labs’ safety claims and whether a model can be shut down reliably.[1] Gillian Hadfield said independent oversight of increasingly powerful and autonomous AI is urgent.[1] The Center Square reported that Dalton was skeptical of the kill-switch concept, arguing that the term is not very meaningful; it also said a function like a kill switch could be used by both good and bad actors.[5]