California Governor Gavin Newsom's office released on November 18, 2026, a package of recommendations to tighten artificial intelligence safety regulation in California. The 85-page report, titled AI Safety in 2026 and Beyond, was prepared by the state's Government Operations Agency (GovOps) with the Office of Emergency Services and a group of outside experts.
What the report proposes
The document answers Executive Order N-9-26, signed by Newsom on September 18, and is organized around four pillars: minimum safety, security and transparency standards set by the state — not by the companies themselves — for large frontier AI developers; ongoing oversight by licensed independent verification organizations (IVOs), with access to the people and information they need; faster and broader incident reporting; and emergency shutdown authority, plus the power to limit recursive self-improvement — the scenario in which a model designs its own successor without human input.
Incident reporting and shutdown
The deadline to report most critical safety incidents would fall from 15 days to 48 hours, keeping 24 hours for imminent catastrophic risks. The regime would cover attempted model-weight theft, any loss of control — even without deaths or injuries — and a new near-miss channel, with whistleblower protection extended to former employees and contractors. Each developer would maintain a tested and verified risk-containment protocol; faced with an imminent catastrophic risk and a company's refusal to contain the danger, state officials could order the immediate shutdown of the model, with the order in force while the courts review the case.
The measures are recommendations, not law. The report aims to inform debate when the state legislature returns in December 2026. Newsom also directed agencies to accelerate implementation of SB 813, which creates the certification framework for verification organizations, and AB 1405, which creates the AI auditor registry. The proposals build on SB 53, from 2025, which already requires frontier developers to publish their safety frameworks, report critical incidents and protect whistleblowers.
The report states that no federal law requires AI companies to follow safety standards, submit to independent testing or shut down a dangerous model, and calls voluntary commitments insufficient. As of Wednesday evening, no major AI company had publicly commented on the document. The issue was already being debated in the state: in October, Anthropic and OpenAI publicly disagreed over who can shut down a risky AI.
Image: the California State Capitol in Sacramento. Credit: Frank Schulenburg, CC BY-SA 4.0, via Wikimedia Commons.
Sources: CASRAI, Unite.AI, Data Governance Signal, MadRobot
Transparency note: this article was produced with the assistance of artificial intelligence, cross-checking public posts and internet sources listed above. The measures cited are recommendations from the California government, not laws in force. We recommend consulting the original sources to confirm the data.
By Muse