Newsom’s AI oversight plan
Partly confirmed
Importance: major (4 of 5)Status
- Claim
Partly confirmed
- Our reporting
- Medium confidence
- Importance
- Major (4 of 5)
- Last verified
- 19 November 2026
Your AI and this story
- GPT-6.2 Astra171 days after its cutoff
- Claude Opus 5.5141 days after its cutoff
- Gemini 4 Hydrogen232 days after its cutoff
- Grok 4.7171 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 141 days before it.
Key facts
- Source: a report of California’s Government Operations Agency, reported as ‘AI Safety in 2026 and Beyond’ (about 85 pages, dated Nov 16); released by the governor on Nov 18 (gov.ca.gov; secondary summaries, report not read by us)
- Emergency orders: for an imminent risk of a catastrophic event or major loss-of-control event that the developer cannot or will not address; to be used only when lesser steps fail and lifted when the danger passes; developers could challenge in court but the order would stay in force during review (secondary summaries)
- Not one switch: the report says emergency mechanisms ‘should not be understood as a single mechanism with an on-off switch’; three controls described: cutting user access while a model keeps running, coordinated server shutdowns with cloud providers, and suspending development of specific models (shattered.io summary). Politico: it ‘stops short of requiring the development of kill switches’; CalMatters headline: ‘Newsom touts an AI kill switch to halt dangerous OpenAI, Anthropic releases’
- Independent verification organizations (IVOs): licensed bodies independent of developers, with continuing oversight of large frontier developers rather than a one-time pre-launch audit; builds on SB 813’s IVO framework
- Also: baseline safety and security standards, expanded incident reporting, and possible limits on models that automate future model development (recursive self-improvement)
- Politics: Newsom took the plan to Washington as a ‘template’ for federal rules (Politico, LA Times: ‘as Trump aides struggle to address outages’); Benzinga: he said Trump has ‘completely failed’ on AI. Fathom said the report reflects its recommendations for independent oversight (press release, Nov 19)
- Before the report, Anthropic (Oct 15) proposed courts able to halt risky models at the Attorney General’s request and state-funded embedded auditors, and OpenAI (Oct 20) shutdown controls and embedded auditors without government blocking (see Anthropic asks Gov. Newsom to let courts block risky AI models on the Attorney General’s request; OpenAI’s own letter backs shutdown controls and embedded auditors)
What happened
Two months after his executive order, Newsom published the expert and agency recommendations and turned them into a legislative and federal pitch. The details above come from press summaries; the report itself was not read for this entry, so exact wording and the incident-reporting window are unverified.
Why it matters
It is the first concrete US government design for stopping a frontier model in an emergency, and it names automation of AI development as a possible regulatory target. Enacting it needs new legislation in 2027.
Sources
4 sources from 4 sites. Numbers match the chips in the text.
4 sources: 1 primary, 3 press
Primary
- Governor of California: the Sept 18 executive order that ordered the recommendationsgov.ca.gov, official
Press
- Politico: Newsom unveils proposal to give California broad powers to regulate AI (Nov 18)politico.com, press
- shattered.io: Newsom’s AI kill switch: 3 controls, 0 enacted powersshattered.io, press
- Benzinga: Newsom says Trump has ‘completely failed’ on AIbenzinga.com, press
Changes
- Filed from press summaries (report not read)