A kill switch for frontier models, auditors stationed inside frontier labs, and a wider definition of what counts as a safety incident. Those are three of the four areas California wants its own AI oversight laws amended to cover.
Governor Gavin Newsom’s Executive Order N-9-26, signed Friday and effective immediately, orders the Government Operations Agency to convene national experts with the Governor’s Office of Emergency Services and deliver recommendations by November 16, 2026.
The four areas the recommendations must address: independent verification organizations embedded onsite in frontier labs for periodic audits; independent verification of the safety frameworks, transparency reports and risk assessments companies file under state law; a kill switch whose efficacy an outside verifier confirms on an ongoing basis; and an expanded definition of critical safety incidents that would capture the loss-of-control cases large developers have recently reported.
The order also raises the possibility of independent third parties writing safety plans for frontier companies, and describes the shutoff as an emergency mechanism.
Deadlines extend past the recommendations. May 1, 2027 is the target for completing certification framework requirements and publishing application criteria for verification organizations. December 1, 2027 covers the initial requirements for the state registry of AI auditors. The governor’s office called both an acceleration of Senate Bill 813 and Assembly Bill 1405.
California is building the enforcement layer for its own statutes. November will show whether the state stops at disclosure or reaches into deployment.
