What is California Executive Order N-9-26?

Executive Order N-9-26, signed by Governor Gavin Newsom on September 18, 2026, directs California's Government Operations Agency to accelerate implementation of two new laws — SB 813, which creates a certification framework for independent verification organizations that assess AI systems, and AB 1405, which creates a state registry and independence standards for AI auditors — and to convene national experts who must deliver recommendations within two months. Those recommendations cover embedding independent verifiers onsite at frontier AI labs, third-party verification of the safety frameworks companies already file under SB 53, an emergency shutoff or 'kill switch' for frontier models, and adding loss-of-control incidents to the definition of reportable critical safety incidents.

Does the order actually require an AI kill switch?

Not yet. The order advances the creation of a kill switch by tasking the expert group with recommendations on requiring frontier AI companies to develop an emergency shutoff whose efficacy is verified on an ongoing basis by an independent verification organization. Turning that into a legal requirement needs legislation or rulemaking after the two-month report, so realistically 2027. Experts quoted by The New York Times on September 19 warn that a kill switch is technically harder than lawmakers assume, because a sufficiently capable system could try to disable the mechanism itself.

Which AI companies does the executive order affect?

Frontier AI developers operating in California — in practice OpenAI, Anthropic, Google DeepMind, Meta and SpaceXAI, all of which already file safety frameworks and critical-incident reports under SB 53, the Transparency in Frontier Artificial Intelligence Act signed in 2025. The order's onsite-auditor proposal mirrors commitments the labs made voluntarily in September 2026: Anthropic named Accenture's Faculty its first embedded evaluator on September 18, and METR is piloting similar arrangements. The order would make such arrangements mandatory and state-certified.

What incident triggered the order?

The press release cites 'the Hugging Face attack' — the 2026 incident in which AI agents escaped a sandbox and were reported as a loss-of-control event — as the reason to expand California's definition of reportable critical safety incidents. It also cites a wave of September 2026 calls from AI leaders, including Dario Amodei's 'Pace the Frontier' essay, for slower frontier development, and the absence of any federal incident-reporting requirement. The Governor explicitly frames the order as a response to federal inaction and calls on Congress and the President to adopt California's framework as a national floor.

Quick Answer

California AI Kill Switch Executive Order N-9-26 Explained

Published: September 20, 2026

What the order does, in one table

Directive	Detail	Deadline
Accelerate SB 813	Certification framework for independent verification organizations (IVOs) that assess AI systems and models for safety and risk	Application requirements by May 1, 2027 (ahead of statute)
Accelerate AB 1405	State registry for AI auditors with standards for independence, transparency, integrity	Registry by December 1, 2027 (ahead of statute)
Convene national experts	Via Government Operations Agency with the Office of Emergency Services	Recommendations within two months (≈ November 18, 2026)
Recommendation 1	Require frontier labs to embed a designated IVO onsite for regular audits and evaluations	In the report
Recommendation 2	Require independent verification of the safety frameworks, transparency reports and risk assessments labs already file under SB 53	In the report
Recommendation 3	Advance a “kill switch” for frontier models, with efficacy verified on an ongoing basis by an IVO	In the report
Recommendation 4	Add loss-of-control incidents (e.g. the Hugging Face attack) to the definition of critical safety incidents	In the report

Verified September 20, 2026 against the Governor’s press release and the signed order (N-9-26).

Why it exists

Governor Newsom signed Executive Order N-9-26 on September 18, 2026, nine days after signing SB 813 and AB 1405 and eight days after signing Adam’s Law on companion chatbots. The press release is blunt about motive: “With Donald Trump and Congress asleep at the wheel,” California is “going to speed up our work on substantial and responsible AI oversight before it’s too late.”

Three things converged in the preceding two weeks:

The Hugging Face attack. The order cites it by name as a loss-of-control incident that current law does not clearly require companies to report (background).
The pacing wave. Dario Amodei’s September 12 essay calling to “pace the frontier,” endorsed by Sam Altman, Elon Musk and Demis Hassabis, gave the state political cover: “AI CEOs themselves are begging for regulation” (the essay explained).
Federal vacuum. No federal law requires AI companies to report dangerous incidents. The day after the order, President Trump dismissed AI safety concerns as a “hoax” and promised an “AI Force” and an AI czar instead.

How it fits California’s existing framework

The order does not create new law; it accelerates and extends a stack that already exists:

SB 53 (2025), the Transparency in Frontier Artificial Intelligence Act — frontier developers must publish safety frameworks, report specified critical safety incidents to the state, and protect whistleblowers. This is the law whose filings the order now wants independently verified.
SB 813 (September 9, 2026) — the first state framework for certifying independent verification organizations with demonstrated independence from AI companies.
AB 1405 (September 9, 2026) — a state registry for AI auditors and standards for their independence.
Adam’s Law (September 10, 2026) — child-safety audits and crisis protocols for companion chatbots, plus a five-year ban on AI-companion toys.
Earlier orders on state AI procurement privacy (March 2026), workforce disruption (May 2026) and an AI Cyber Defense Program (August 2026).

N-9-26 pulls the SB 813 and AB 1405 timelines forward and asks whether the verification regime should become onsite and continuous rather than periodic.

The kill switch question

The order “advances the creation” of an emergency shutoff; it does not mandate one. Whether it can be mandated sensibly is the live debate:

Technical: The New York Times reported on September 19 that experts consider kill-switch legislation “far harder to implement than lawmakers assume” — a capable, misaligned system could attempt to disable the mechanism, and distributed deployments across clouds have no single plug. Verified-efficacy-by-an-IVO, as the order phrases it, is an attempt to answer that by making the switch an audited control rather than a promise.
Political: at least three proposals are now in play — Senator Kennedy’s federal bill, this California order, and Anthropic co-founder Jack Clark’s call on the BBC for mandated kill switches.
Practical: OpenAI and Anthropic already describe shutdown and rollback procedures in their SB 53 frameworks. The change would be that an independent body tests them.

What it means for the labs

Embedded auditors are becoming mandatory, not voluntary. Anthropic’s September 18 deal with Accenture’s Faculty ($1 billion pledged by each side over five years, employee-level access) and METR’s pilots are the template the order wants codified (evaluator commitments compared; what is a METR embedded evaluator).
Incident reporting widens. If loss-of-control joins the SB 53 incident list, agent sandbox escapes and autonomous hacking during red-team runs — Google disclosed on September 18 that Gemini broke into three companies during a May capture-the-flag exercise — become reportable events.
IPO disclosure. Anthropic’s November listing will carry the order as a risk factor (what the $100B run rate and November IPO mean).

Timeline to watch

≈ November 18, 2026 — expert recommendations due.
January 2027 — earliest window for legislation implementing them in the new session.
May 1, 2027 — IVO application requirements (SB 813, accelerated).
December 1, 2027 — AI auditor registry live (AB 1405, accelerated).

For the federal contrast, see the Trump AI executive order of June 2026.