Regulatory  ·  2026-11-21

California releases recommendations for mandatory AI safety standards, third-party oversight, emergency shutdown authority and RSI caps

RegulatoryHigh impactUnited States
On 18 November 2026, Governor Gavin Newsom released recommendations (developed by an expert group convened under his executive order and the Government Operations Agency) to strengthen California's AI safety framework. The recommendations propose: state-set baseline safety standards for all AI developers with independent safety testing/verification across model training, development and deployment; expanded safety-incident reporting and verified risk-containment/shutdown protocols giving the state authority to order an immediate emergency shutdown; and authority to cap the development and use of recursive self-improvement (RSI). The Governor also ordered state agencies to fast-track SB 813 (independent verification organisations) and AB 1405 (state registry of AI auditors).
This is authoritative state action pushing toward a legally binding, government-enforced AI-safety regime — mandatory oversight, state shutdown powers and RSI limits — beyond the voluntary federal approach. It will shape compliance for any organisation training or deploying frontier models in or serving California, and sets a template other states may follow. Independent evaluation, incident reporting and shutdown protocols create concrete engineering and governance obligations.
Frontier model developers operating in California should prepare for mandatory independent safety testing, incident-reporting channels, documented risk-containment/shutdown protocols, and RSI capping; CDOs/GCs should map model lifecycle governance to these requirements now, as enabling legislation is being fast-tracked.
Office of Governor Gavin NewsomCalifornia Government Operations Agency — report (PDF)
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →