California Governor Newsom proposes mandatory oversight of frontier AI and granting state officials shutdown power
Newsom schlägt verpflichtende Aufsicht über Frontier-KI und Befugnisse zur Abschaltung vor
California Governor Newsom issued legislative recommendations on November 18, 2026, proposing continuous independent oversight of large frontier AI model developers and authorizing state officials to order immediate shutdown when a model poses an imminent catastrophic danger or serious loss of control and the developer refuses to contain it. The recommendations are based on the report "AI Safety 2026 and Beyond" released by the California Government Operations Agency on November 16, for legislators' reference when they reconvene in December, and are currently non-binding. Developers must undergo continuous audits by accredited independent verification organizations (IVOs), maintain risk mitigation protocols covering compute, agents, internal access and deployment, and test shutdown failure scenarios such as model weight replication; open-weight developers' obligations are limited to deployments they control. Most key incident reporting deadlines are proposed to be shortened from 15 days to 48 hours; attempted model weight theft, loss of control and attempted incidents must all be reported, and models capable of causing catastrophic danger must be reported as frontier models even if training compute is below 10^26. Automated AI R&D companies must also report compute allocation quarterly, and a preventive cap subject to periodic review is proposed for recursive self-improvement (RSI).