California Governor's Expert Report Urges AI Kill Switch, Mandatory Standards and RSI Compute Caps
California Governor Gavin Newsom announced on November 19 a strengthened AI safety framework for the state. GovOps, together with Cal OES and four experts—Alondra Nelson (Princeton), Jason Goldman, Robert Reich (Stanford), and Gillian Hadfield (Johns Hopkins)—released the report "AI Safety in 2026 and Beyond," putting forward a set of mandatory regulatory recommendations that have been widely read as a call for urgent action.
Confirmed
- Emergency shutdown authority: the government could forcibly shut down dangerous AI systems
- Baseline safety standards: applying to all AI developers, set by the state rather than by companies themselves
- Mandatory ongoing third-party oversight: for advanced AI development
- Compute limits on RSI: caps on compute tied to recursive self-improvement
- The report was officially released by the Governor's office and represents the next step in California's AI safety governance
Why it matters
- This is a governor-led, mandate-oriented AI safety plan, marking California's shift from voluntary commitments to mandatory regulation
- If implemented, the kill switch and compute caps would directly constrain frontier AI companies' R&D and deployment—especially significant as a model in the absence of federal regulation
2026-11-19 ~ 2026-11-19 · 8 related posts
Primary sources
- Newsom's AI expert panel calls for emergency shutdown powers, safety standards, compute caps — trevposts · 2026-11-19
- [source] California moves to add emergency shutdown authority and cap recursive self-improvement in AI — ShakeelHashim · 2026-11-19
- California Report Backs Mandatory AI Baseline Standards, Shutdown Powers, and RSI Caps — sjgadler · 2026-11-19
- California expert panel calls for emergency AI shutdown authority and mandatory oversight — Miles_Brundage · 2026-11-19
- California AI safety report proposes compute-disclosure rules and RSI caps — sethlazar · 2026-11-19
- Newsom's AI policy report calls for mandatory evaluators and limits on recursive self-improvement — mattsheehan88 · 2026-11-19
2 near-duplicate retellings: Manderljung · Miles_Brundage