California Proposes Mandatory AI Safety Rules and State Shutdown Power
The November 18 report asks lawmakers to require state-set standards with compulsory third-party verification plus limits on self‑improving systems as a blueprint for federal action.
Overview
- Gov. Gavin Newsom's office released the expert-informed recommendations on November 18, 2026 calling for the state to set baseline safety rules and require independent testing at every stage of model development.
- The plan would expand incident reporting, force firms to create verified risk‑containment and shutdown protocols, and give California officials the authority to order immediate shutdowns of dangerous systems.
- It seeks the legal power to cap recursive self‑improvement, defined as an AI designing a more capable successor, to curb scenarios that could lead to rapid, uncontrolled gains in capability.
- The recommendations are advisory and would need new state laws or regulations to take effect, and Newsom is urging both the California Legislature and federal leaders to adopt the framework nationally.
- Coverage presents the package as both a possible national template and political positioning for Newsom, and it builds on California's multi‑year AI laws dating to 2023 that already require disclosures, audits, and incident reporting.