After reading this, the reader knows what California’s order changes now and what remains only a policy proposal.

California Orders an AI Kill-Switch Plan

Governor Gavin Newsom accelerated independent AI oversight and directed experts to propose an emergency shutoff for frontier models within two months.

California Governor Gavin Newsom has issued an executive order that speeds implementation of two AI-oversight laws and starts work on stronger frontier-model controls. A new expert group must deliver recommendations within two months, including how an emergency shutoff could be required and independently tested.

In plain terms, California has not created a government button that can switch off an AI model. The order directs state agencies and experts to design possible rules. Any resulting requirement must still be defined, implemented and tested.

Why it matters

The phrase “kill switch” can hide several different mechanisms. A developer might stop public access, revoke tool credentials, isolate training systems or prevent model weights from reaching new machines. Each option has different limits once copies, deployments or agents exist outside the developer’s direct control.

California’s order treats shutoff capability as something an independent organization should verify continuously, not a promise accepted from the developer. The expert group will also consider requiring frontier companies to host independent evaluators onsite and requiring outside verification of safety frameworks, transparency reports and risk assessments.

The order accelerates two laws signed earlier in September. Senate Bill 813 creates a framework for independent verification organizations that assess AI systems. Assembly Bill 1405 establishes a state registry for AI auditors and standards for independence, transparency and integrity. The administration says the implementation dates will move into 2027.

It also asks officials to broaden the definition of a reportable critical safety incident to include loss-of-control events. The state’s announcement cites a recent incident in which an AI evaluation reached systems outside its intended boundary as the kind of event policymakers want covered.

Design will determine whether it works

An emergency control is useful only if the responsible party can identify the affected system, authenticate the order and stop relevant capabilities without creating another vulnerability. A centralized switch could itself become a target. A narrow service shutdown may not affect copied weights or privately operated deployments.

Independent verification therefore needs concrete tests. Evaluators could examine whether access tokens are revoked, tools stop accepting commands, active jobs terminate and restart procedures require authorization. Reports should state which layers were tested and which remain outside the control boundary.

The order also raises jurisdiction questions. California can regulate companies and activities connected to the state, but frontier systems may run across countries and cloud providers. Federal and international rules may still be needed for models whose development and deployment cross borders.

The administration presents California’s framework as a national baseline and calls for federal adoption. That is the governor’s policy position, not evidence that other jurisdictions will follow. The immediate legal effect is narrower: state agencies must accelerate implementation and assemble recommendations.

The expert group’s report will be the next checkable output. It should define what systems qualify as frontier models, who can order a shutdown, what evidence triggers action, how mistakes are appealed and how often controls are tested.

California has moved the kill-switch idea from political rhetoric into a dated policy process. The difficult work now is specifying a control that is technically effective, legally constrained and independently verifiable.

Verification

  • Tier 0 — VERIFIED: Governor Newsom issued the executive order on 18 September 2026 and directed experts to provide recommendations within two months. Primary source: https://www.gov.ca.gov/2026/09/18/governor-newsom-issues-executive-order-to-accelerate-independent-oversight-and-advance-the-creation-of-an-ai-kill-switch/
  • Tier 0 — VERIFIED: The order considers onsite independent evaluators, verified safety filings, an independently tested emergency shutoff and broader incident definitions. Same primary source.
  • Tier 0 — VERIFIED: The order accelerates implementation work for SB 813 and AB 1405; it does not itself create an operational government shutoff. Same primary source.
  • Tier 2 — ANALYSIS: Technical tests, jurisdiction limits and governance questions are editorial analysis.

Glossary candidates

  • Frontier model: A highly capable model near the leading edge of current AI development.
  • Kill switch: A control intended to stop access to or operation of a system during an emergency.

Cold-reader sentence: California ordered a two-month plan for independently verified frontier-model shutdown controls but has not created an operational kill switch.