California released recommendations on Nov. 18 that would give state officials power to order an immediate emergency shutdown of frontier AI systems they deem dangerous, require developers to prove they can shut down their own models, and cut the deadline for reporting critical AI safety incidents from 15 days to 48 hours. Gov. Gavin Newsom touted the plan as a blueprint for state legislation.
The recommendations come from a report titled AI Safety in 2026 and Beyond, produced by the California Government Operations Agency with the Office of Emergency Services and outside advisers Jason Goldman, Gillian Hadfield, Alondra Nelson and Rob Reich. It responds to a Sept. 18 executive order in which Newsom asked for proposals on independent oversight and an AI kill switch within two months. The report covers catastrophic risk only, and none of its proposals are law yet.
The core of the plan is continuous oversight. Every large frontier developer would be monitored by independent verification organizations licensed by the Government Operations Agency, which could revoke those licenses. The state, not the AI companies, would assign evaluations to auditors. Auditors would get access to models, personnel, systems, training environments and records, and would send unredacted assessments to the state.
Key facts
- 01California's 'AI Safety in 2026 and Beyond' report was published Nov. 18, 2026, by the Government Operations Agency.
- 02It proposes letting state officials order an immediate emergency shutdown of frontier AI systems facing imminent catastrophic risk.
- 03Critical AI safety incident reports would be due in 48 hours instead of 15 days.
- 04Large frontier developers would face ongoing oversight by state-licensed independent verification organizations.
- 05Lawmakers return to Sacramento on Dec. 7, when bills based on the recommendations can be introduced.
Each large frontier developer would also maintain a Risk Containment Protocol covering compute, agents, internal access and deployment, verified by an auditor and certified by a senior company official. Testing would include failure cases such as a model copying its own weights to resist containment. Open-weight developers would be responsible only for deployments they control.
The emergency shutdown order would apply when officials have credible reason to believe there is an imminent catastrophic risk or major loss-of-control event and the developer refuses to contain it. The power could apply to companies such as OpenAI and Anthropic. The order would have to be the narrowest workable intervention, would take effect immediately and stay in force during judicial review, and developers could seek a temporary restraining order.
Incident reporting would expand to cover attempted theft of model weights, all loss-of-control incidents and a new near-miss channel, while keeping the existing 24-hour deadline for imminent catastrophic risk. Whistleblower protections would extend to contractors and former employees. Models capable of actions that could create catastrophic risk would count as frontier models for reporting purposes even if trained with fewer than 10^26 operations.
Companies automating AI research would report each quarter how their compute splits among training, research experiments, AI research agents, safety work and external customers. The report also proposes standby cap authority, letting officials impose a time-limited cap on automated AI research if it outpaces safeguards. A surcharge on auditor fees would fund CalCompute, the state's public cloud initiative, with part of that capacity reserved for independent safety research.
The plan builds on California's existing requirement that large frontier developers publish safety frameworks. State law defines catastrophic risk as a single event capable of killing more than 50 people or causing $1 billion in theft or property damage. Newsom also accelerated implementation of laws creating an AI auditor registry and auditor standards, and framed the effort as filling a gap left by the absence of federal AI rules.
Anthropic said it was encouraged to see the report move the conversation forward on independent evaluation of AI models, according to Tino Cuéllar, its chief global affairs officer. No reaction from OpenAI was reported.
“We will continue working with leaders in California and throughout the world to make sure we keep making progress”— Tino Cuéllar, Chief global affairs officer, Anthropic
Governor-elect Xavier Becerra said in a written statement that he will move quickly with the Legislature to put full incident reporting and real shutdown authority into law. Lawmakers return to Sacramento on Dec. 7, when new bills can be introduced.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.




