Newsom Wants AI Shutdown Power; Incident Reports in 48 Hours

California Governor Gavin Newsom on 18 November released recommendations to strengthen the state’s AI safety law, SB 53, including state authority to order an emergency shutdown of a dangerous model, mandatory independent oversight and power to cap recursive self-improvement, according to his office.

Our number: the report would cut the deadline for reporting most critical safety incidents from 15 days to 48 hours, a window 7.5 times shorter by our arithmetic (360 hours to 48), while keeping the 24-hour rule for imminent catastrophic risks.

Business Pill · Least Privilege

An evergreen Business Engineer explainer on least privilege and access control. It teaches the general idea only and does not discuss California’s report.

The key insight: Newsom’s report would let California order an immediate emergency shutdown of a dangerous model, cap recursive self-improvement and cut most incident reporting from 15 days to 48 hours, 7.5 times shorter by our arithmetic.

What the Report Recommends

The 85-page report, “AI Safety in 2026 and Beyond”, from the California Government Operations Agency, is dated 16 November and was written in consultation with the California Office of Emergency Services and four outside experts. It answers an executive order that asked for recommendations in four areas: standards, independent verification organizations, critical safety incident reporting and emergency shutdown requirements.

On standards, it says “The state should establish baseline standards for safety, security, and transparency that all large frontier AI developers must follow”. On automated AI research, it says developers should report “the share of compute they are devoting to this purpose”, and that the state should have authority to cap the development and use of RSI if it judges the risk catastrophic.

The report would cut the deadline for reporting most critical safety incidents from 15 days (360 hours) to 48
The report would cut the deadline for reporting most critical safety incidents from 15 days (360 hours) to 48 hours, 7.5 times shorter by our arithmetic.

Shutdown Orders, Not a Kill Switch

The report proposes that every large frontier developer keep a tested Risk Containment Protocol. It says: “If the danger from a model is imminent and the developer refuses to take action to contain the danger, the state should have authority to order an immediate emergency shutdown”. The developer could seek a temporary restraining order, but the order would take effect immediately.

Politico reports that the plan “stopped short of calling for the development of so-called kill switches”. The report itself says the emergency mechanisms “should not be understood as a single mechanism with an on-off switch”.

Who Would Be Covered

The report says relying only on the 10^26-operation compute threshold “has prevented the state from scoping in dangerous incidents”, and proposes covering models with catastrophic risk even below it. It also suggests widening the definition of a large frontier developer beyond those with annual gross revenues of more than $500 million.

The Politics

Newsom framed the plan as a model for Congress. Politico reports he released it in Washington, called it “an action document”, and that the recommendations are not guaranteed to become law; Gov.-elect Xavier Becerra and the Legislature would decide. Politico also reports that OpenAI, Anthropic and Meta sent the governor’s office policy proposals. Anthropic earlier asked Newsom to let courts suspend risky models, as we reported; for the safety debate around Anthropic this week, see the MisAnthropic Statement.

The Structural Read

As we read the report, the biggest change is the RSI cap: it makes the share of compute spent on automated AI research a reported, verifiable metric.

As we read it, the shutdown power is designed for a refusing developer, with court review after the order rather than before it.

Gavin Newsom, Governor of California (via Politico)

“This is an action document”

Three Implications

FOR FRONTIER LABS The report would require a tested Risk Containment Protocol and independent verification across training and deployment.

FOR AI AUDITORS Independent verification organizations would verify RSI metrics first, under SB 813 and AB 1405.

FOR LAWMAKERS These are recommendations; the Legislature and Gov.-elect Becerra would decide what becomes law.

What Is Not Established

These are recommendations, not bills. No legislative text, penalties or effective dates are set yet, and the RSI cap level is not defined in the report. We did not contact the governor’s office or AI companies for comment.

This Week In The AI Supercycle

AI Lost Its Central Banker, the Lab Oligopoly, and the Future of Open Source

This week’s issue: the Nvidia leadership change, read as the AI build-out losing its central banker: who now finances the GPU landlords, backstops the data centres and sponsors open models, and why two labs are becoming the bank.

Read the full issue →

Business Engineer Framework

Inside Anthropic’s Permission Layer

A Business Engineer analysis of who sets the rules, permissions and boundaries around frontier AI.

Read the Map of AI →

The Bottom Line

Newsom’s report would give California emergency shutdown power over dangerous models, a cap on recursive self-improvement and a 48-hour incident-reporting deadline, 7.5 times shorter than today’s 15 days by our arithmetic.

97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. The recommendations and quotes are from the governor’s 18 November release and the GovOps report, read in full; the kill-switch framing, Washington event and company submissions are attributed to Politico. The 7.5 times figure is our arithmetic (15 days = 360 hours, divided by 48). We did not contact the governor’s office or AI companies.

Sources: Office of the Governor of California, 18 Nov 2026 · California Government Operations Agency, AI Safety in 2026 and Beyond, 16 Nov 2026 · Politico, 18 Nov 2026

Scroll to Top
FourWeekMBA