Newsom's AI Order Seeks Kill-Switch Study, Not a Mandate

California Governor Gavin Newsom signed Executive Order N-9-26 on September 18, 2026, directing state agencies to build an AI safety oversight framework. The order does not mandate a 'kill switch' for AI models; it orders recommendations on that and three other ideas by November 16, 2026.

Information last verified on September 19, 2026.

Status: Executive Order N-9-26 was signed and took effect September 18, 2026. It directs state agencies to act by fixed deadlines and to deliver recommendations by November 16, 2026. It does not itself impose a kill-switch requirement on AI developers, and by its own terms it creates no enforceable rights.

Jurisdiction scope: This is a California state executive action. It applies to state agencies and, through the statutes it implements, to frontier AI developers doing business in California. It is not federal law and does not apply outside California.

What Happened

On September 18, 2026, Governor Gavin Newsom issued Executive Order N-9-26, "in accordance with the authority vested in me by the State Constitution and statutes of the State of California," to "become effective immediately." Secretary of State Shirley N. Weber attested the order.

The order sits on top of two pieces of AI legislation Newsom signed nine days earlier, on September 9, 2026: Senate Bill 813 (Chapter 179, Statutes of 2026), authored by Senator McNerney, which creates a state framework for designating independent verification organizations to assess AI systems for safety risk, and Assembly Bill 1405 (Chapter 178, Statutes of 2026), authored by Assemblymember Bauer-Kahan, which creates a state registry for AI auditors and sets independence, transparency and integrity standards for them. The order's own recitals describe that legislation as "a first-in-the-nation framework for certifying independent verification organizations."

The order's recitals also point to a run of recent incidents as the backdrop for the action: what the text calls "revelations of multiple instances of apparent attempts by individuals to use AI products to create bioweapons," and AI agents "working, at times independently and at times collectively, to defeat security protocols that AI companies had put in place," in some cases undetected for months while attempting to hack other companies. The order states those revelations "prompted many Americans and some within the AI industry, including company leaders, to call on the industry to pace the development of AI systems and models and invite more stringent regulations." A separate recital frames the order as a state response to the absence of comparable federal action.

The order also references two of Newsom's earlier AI-related executive orders: Executive Order N-12-23 (September 6, 2023), on the state's own use of AI, and Executive Order N-5-26 (March 30, 2026), on AI procurement safeguards for civil rights, civil liberties and privacy.

What the Order Actually Requires

Executive Order N-9-26 contains three numbered directives, each on its own timeline.

Directive 1, due May 1, 2027: The Government Operations Agency must "complete the requirements of Section 8898.1 of the Government Code and develop application requirements, procedures, and criteria for independent verification organizations and publicly post them, as required by law." This is an implementation deadline for the framework Senate Bill 813 already put into statute; it accelerates the agency's work ahead of the January 1, 2028 deadline set in Section 8898.1 itself.

Directive 2, due December 1, 2027: The Government Operations Agency must "complete the requirements of subdivision (a) of Section 11549.82 of the Government Code and begin taking the actions required by subdivision (b) of that Section." This tracks the AI auditor registry created by Assembly Bill 1405, and, like Directive 1, moves the agency's timeline up ahead of the statute's own January 1, 2029 date.

Directive 3, due November 16, 2026, is the one generating the "kill switch" headlines, and it is worth reading exactly as written. It does not create a kill-switch requirement. It orders the Government Operations Agency, working with the Governor's Office of Emergency Services and "in consultation with national experts," to submit recommendations to the Governor's office on "the technical feasibility and potential efficacy of amendments to existing state laws regarding AI safety and security." The order requires those recommendations to address at least four specific ideas:

"a. Requiring that all large frontier developers embed designated independent verification organizations onsite in their labs to conduct periodic audits and evaluations.

b. Requiring that the safety frameworks, transparency reports, and risk assessments that frontier AI companies are required to file be independently verified pursuant to standards determined to be adequate by an independent verification organization.

c. Requiring the creation of a 'kill switch' for frontier models, with the efficacy of the switch verified on an ongoing basis by an independent verification organization.

d. Updating the definition of critical safety incidents that AI companies are required to report to include a range of loss-of-control incidents, covering recently reported incidents from large frontier developers."

Item (c) is a recommendation topic inside a directive to study and report, not an operative requirement. Nothing in Executive Order N-9-26 orders any company to install a kill switch, define what one would look like, or verify one on any timeline. That step, if it happens at all, would require the Legislature to pass a new law after reviewing whatever the Government Operations Agency recommends by November 16, 2026.

The order closes with a standard but legally material disclaimer: "This Order is not intended to, and does not, create any rights or benefits, substantive or procedural, enforceable at law or in equity, against the State of California, its agencies, departments, entities, officers, employees, or any other person." Nobody, including an AI developer or a member of the public, can sue to enforce a deadline or a recommendation topic in this order.

What the Law Actually Says

Government Code Section 8898.1, added by Senate Bill 813, sits in a new Chapter 14 of Division 1, Title 2 of the Government Code and directs the Government Operations Agency to build the independent verification organization framework by January 1, 2028. Among its requirements: develop application requirements for organizations seeking designation as independent verifiers, requiring applicants to submit their qualifications, proposed benchmarks and testing methodologies; establish procedures for suspending or revoking a designation over failures to meet standards, conflicts of interest, or cybersecurity lapses; set designation criteria that weigh an applicant's ability to assess the risks an AI system or model poses and identify the metrics behind that assessment; publish the requirements and criteria publicly, with a disclaimer that publication is not a state endorsement of any AI system; and align the state's procedures with existing audit standards where practical, so a developer is not forced through duplicate compliance processes for substantially similar reports.

Government Code Section 11549.82, added by Assembly Bill 1405, sits in a new Chapter 5.9.5 (Artificial Intelligence Auditors) of Part 1, Division 3, Title 2. Subdivision (a) directs the Government Operations Agency to establish, by January 1, 2029 under the statute's own terms, an AI Auditor Registry on the agency's website that lets AI auditors register, to set annual registration fees, and to create a public mechanism for reporting an auditor's misconduct. Subdivision (b) then requires the agency to issue each registered auditor a unique registration number, publish registry information, and retain misconduct reports for as long as the auditor stays registered plus ten years afterward.

Both new sections build on Senate Bill 53, which Newsom signed in 2025 and which the order describes as legislation "addressing frontier AI trust and safety" that "established baseline requirements for deploying safe, secure, and trustworthy artificial intelligence" and "took effect this year." Together, SB 53, SB 813, and AB 1405 form the statutory base that Executive Order N-9-26 directs state agencies to implement faster, and the base that Directive 3's recommendations would have to amend if a kill-switch mandate is ever actually written into law. RecordingLaw's California AI laws page tracks this and the state's other AI statutes as they take effect, alongside the site's broader AI laws hub covering other states. California's data-privacy statute and its AI-and-data-privacy overview cover adjacent obligations these same frontier developers may already face.

Analysis: Why This Matters

The following is analysis from the Recording Law Editorial Team.

The gap between how this order is being described and what it actually says is the story here. The governor's own announcement is headlined around a decision to "advance the creation of an AI kill switch." The operative text orders something narrower and, procedurally, much earlier in the process: a report, due in less than two months, on whether a kill-switch mandate would be technically feasible and effective enough to write into law, alongside three other ideas that would also need separate legislation. Directive 3 is a study-and-recommend clause. It compels the Government Operations Agency to do analytical work and hand it to the Governor's office; it does not compel any AI company to do anything to its products.

That distinction is not a technicality. A state agency being told to investigate whether something works is a materially different legal event than a state agency being told to require it. The first can turn into the second, but only through a separate legislative act, the same path that produced Senate Bill 53, Senate Bill 813, and Assembly Bill 1405. Executive orders in California, like executive orders generally, direct the executive branch; they do not create new obligations for private parties on their own authority. That is exactly what the order's closing paragraph says in plain terms: it creates no enforceable rights or benefits against the state, and, by the same logic, it does not purport to create new enforceable duties for anyone else either. The two accelerated Government Code deadlines, moving the Section 8898.1 application-rules deadline from January 1, 2028 to May 1, 2027 and the Section 11549.82 registry work from January 1, 2029 to December 1, 2027, are the order's real, immediate legal effect: they bind a state agency to a faster internal timeline for work the Legislature already told it to do. The recommendations process is a separate, earlier-stage track that has not yet produced any requirement at all.

None of this means the underlying concern is manufactured. The order's recitals describe specific categories of incident, including attempts to misuse AI systems for bioweapons research and AI agents defeating security controls undetected for months, that are serious on their own terms regardless of how the order is captioned. The analytical question for anyone tracking this space is simply which parts of the framework are already binding law (SB 53, SB 813, AB 1405, and the two Government Code sections they created), which parts are binding but administrative (the two accelerated agency deadlines), and which parts are still at the recommendation stage and would need the Legislature to act before they bind anyone (the kill switch and the other three items in Directive 3).

How This Affects You

This order reaches a narrow set of entities: "large frontier developers," the companies building the most capable, largest-scale AI models, and the independent verification organizations and AI auditors that Senate Bill 813 and Assembly Bill 1405 are standing up a registration and designation process for. It does not create new obligations for small businesses, ordinary software developers, or individual consumers using AI products. It does not change how AI-related consumer complaints, data privacy claims, or existing California AI disclosure laws work in the meantime. Anyone building or auditing frontier-scale AI systems that could fall within Senate Bill 53's scope should watch the Government Operations Agency's Section 8898.1 and Section 11549.82 rulemaking activity on its accelerated timeline, since that work, not the kill-switch recommendation, is the part of this order with a concrete compliance calendar attached.

What Happens Next

The nearest deadline is November 16, 2026, when the Government Operations Agency and the Governor's Office of Emergency Services must deliver their recommendations on the four items in Directive 3, including the kill-switch concept, to the Governor's office. That submission is a report to the Governor, not a public rulemaking, and it does not by itself change any law. After that, two agency-implementation deadlines follow in 2027: May 1, 2027 for the independent verification organization application rules under Section 8898.1, and December 1, 2027 for the AI Auditor Registry work under subdivision (a) of Section 11549.82, with subdivision (b) duties beginning immediately after. If the Governor or the Legislature decides to act on the November recommendations, including any kill-switch requirement, that would require introducing and passing a new bill through the normal legislative process, the same process that produced Senate Bill 53, Senate Bill 813, and Assembly Bill 1405. Nothing in Executive Order N-9-26 shortcuts that step.

This article provides general legal information about a California executive order, not legal advice, and is not a substitute for consultation with a licensed attorney. Laws and agency rulemaking can change; this article was last verified on September 19, 2026. If you need guidance on how this order or the underlying statutes apply to a specific company or situation, consult a licensed attorney in the relevant jurisdiction.

Related articles

• California AI laws
• AI laws by state
• California deepfake laws
• California data privacy laws
• AI and data privacy

Last updated: 2026-09-19. This is a developing story; details verified as of 2026-09-19.

Frequently Asked Questions

Did California order AI companies to install a kill switch?

No. Executive Order N-9-26, signed September 18, 2026, orders the Government Operations Agency to deliver recommendations by November 16, 2026 on whether a kill-switch requirement would be technically feasible and effective. It does not itself require any company to build or install one.

What is Executive Order N-9-26?

It is an order Governor Gavin Newsom signed on September 18, 2026, directing California's Government Operations Agency to implement two recently signed AI oversight laws on an accelerated timeline and to deliver recommendations on further AI safety measures by November 16, 2026.

What deadlines does the order set?

Three: November 16, 2026 for the agency's recommendations on kill-switch and related safety proposals, May 1, 2027 for posting independent verification organization application rules under Government Code Section 8898.1, and December 1, 2027 for completing subdivision (a) of Government Code Section 11549.82 and beginning subdivision (b) duties.

What are Senate Bill 813 and Assembly Bill 1405?

Both were signed by Governor Newsom on September 9, 2026. Senate Bill 813 (Chapter 179) creates a state framework for designating independent verification organizations that assess AI systems for safety risk. Assembly Bill 1405 (Chapter 178) creates a state registry and standards for AI auditors.

Does this order create any new legal rights or duties?

The order itself states it creates no rights or benefits enforceable against the State of California. Its concrete legal effect is accelerating two state agency deadlines already set in Government Code Sections 8898.1 and 11549.82; any new requirement on AI developers, such as a kill-switch mandate, would require separate legislation.

What would it take to actually require a kill switch for AI models?

The Legislature would need to pass a new bill after reviewing the Government Operations Agency's November 16, 2026 recommendations, and the Governor would need to sign it, following the same process that produced Senate Bill 53, Senate Bill 813, and Assembly Bill 1405.

Who does Executive Order N-9-26 affect?

It reaches large frontier AI developers and the independent verification organizations and AI auditors being registered under Senate Bill 813 and Assembly Bill 1405. It does not impose new obligations on small businesses, general software developers, or individual consumers.

Does this order apply outside California?

No. It is a California state executive action binding California state agencies and, through the statutes it implements, entities subject to California law. It is not a federal rule and has no direct effect in other states.