Executive Order N-9-26 directs the Government Operations Agency to finish California's independent verification organization and AI auditor registry set-up months before the statutory dates and to submit recommendations on onsite auditors and a frontier-model kill switch by November 16, 2026

Three dates, and what each one moves

The order has three numbered paragraphs and three dates. Paragraph 1 directs the Government Operations Agency, no later than May 1, 2027, to complete the requirements of Government Code section 8898.1 and to develop application requirements, procedures and criteria for independent verification organizations and publicly post them. Paragraph 2 directs the same agency, no later than December 1, 2027, to complete the requirements of subdivision (a) of section 11549.82 and to begin taking the actions required by subdivision (b). Paragraph 3 directs the agency, in consultation with the Governor's Office of Emergency Services, to submit recommendations to the Governor's office no later than November 16, 2026.

Section 8898.1, as enacted by SB 813 and approved on September 9, 2026, gives the agency until January 1, 2028 for the same work. Section 11549.82(a), as enacted by AB 1405 on the same day, gives it until January 1, 2029, and subdivision (b) begins on that same date. So the order moves the agency's own clock forward by eight months on the first regime and by thirteen months on the second. That arithmetic is ours; the order states calendar dates and does not describe them as accelerations.

Note the verb in paragraph 2. The agency is to complete subdivision (a) and begin subdivision (b) by December 1, 2027. Subdivision (b) is a list of continuing actions, including issuing registration numbers, publishing registrant information and retaining misconduct reports. The order does not say those are to be finished by December 2027, and we do not read it that way.

Two regimes, not one

Paragraphs 1 and 2 read, at first pass, like a single auditor scheme. They are not. Section 8898.1 belongs to SB 813, under which the agency designates independent verification organizations, the entities the statute calls IVOs, on application, against published criteria, with procedures for suspending or terminating a designation. Section 11549.82 belongs to AB 1405, under which the agency runs an AI Auditor Registry, fixes annual registration fees and maintains a mechanism for reporting auditor misconduct. Designation and registration are different acts with different consequences, and the order keeps them in separate paragraphs.

What the order does not move is the private-party date in AB 1405. Section 11549.82.5 provides that beginning January 1, 2029, a person shall not offer, sell or conduct a covered AI audit unless registered with the agency. Paragraph 2 of the order names section 11549.82 and nothing else. An auditor's registration obligation still commences on the statutory date; what changes is when the registry that would receive the registration is expected to exist.

For a prospective IVO the practical consequence is the earlier date. The order brings the deadline for completing the section 8898.1 work and posting the application materials forward by eight months, from January 1, 2028 to May 1, 2027. It does not specify when applications will open, and the statute permitted earlier action. Whether the agency meets that date is not something the order can guarantee, and we make no prediction about it.

Four recommendation topics, and the word that governs them

Paragraph 3 is the part of the order most easily overstated. The agency is directed to submit recommendations, developed in consultation with national experts, "addressing the technical feasibility and potential efficacy of amendments to existing state laws regarding AI safety and security". The recommendations are to include at least four subjects, listed at 3(a) to 3(d).

Each of the four is a topic on which paragraph 3 asks for recommendations; none of them is a requirement the order imposes, and the order says nothing about whether any will be adopted. Subparagraph (a) is embedding designated independent verification organizations onsite in the labs of all large frontier developers to conduct periodic audits and evaluations. Subparagraph (b) is independent verification of the safety frameworks, transparency reports and risk assessments that frontier AI companies are required to file, against standards an IVO determines to be adequate. Subparagraph (c) is the creation of a "kill switch" for frontier models, with the efficacy of the switch verified on an ongoing basis by an IVO; the order sets the term in typographic quotation marks, which we render as straight quotes here. Subparagraph (d) is updating the definition of critical safety incidents that AI companies are required to report so that it covers a range of loss-of-control incidents, including recently reported incidents from large frontier developers.

Each of the four is introduced by a participle: requiring, requiring, requiring, updating. None of them is a requirement. They describe what the recommendations are to address, and the recommendations themselves are to address feasibility and potential efficacy, which is the language of a study, not of a rule. The order does not say which statutes would be amended. It refers to amendments to existing state laws regarding AI safety and security generally; SB 53, SB 813 and AB 1405 are not named in paragraph 3, and we do not attribute to the order any plan to amend any of them. The recitals do describe the exercise as developing a roadmap for further strengthening the state's laws, and that is the Governor's characterisation, not ours.

The report is due November 16, 2026. It goes to the Governor's office. The order does not say whether it will be published, and does not set any step after delivery.

What the order does not do

It does not designate any IVO, register any auditor, certify any lab, install any kill switch or change the definition of a reportable critical safety incident. Every one of those is either an agency action still to be taken under the two statutes or a topic for the November report.

It creates no private right. The paragraph after the ordering paragraphs states that the 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". A developer, an auditor and a member of the public are all outside the order's enforcement reach.

The recitals describe the backdrop as revelations of apparent attempts by individuals to use AI products to create bioweapons, and of AI agents working to defeat the security protocols of AI companies and, in some instances undetected for months, to hack other companies. Those are the Governor's stated reasons. The order cites no source for them and we did not investigate them; we report them as the rationale the instrument gives for itself, and not as established events.

The order also directs that it be filed with the Secretary of State as soon as possible and given widespread publicity. That is an instruction, not a record of filing. The PDF carries an ATTEST line for the Secretary of State beneath the Governor's signature block; that is part of the order's execution, not a filing receipt.

What we did not verify

What we opened: the signed three-page PDF of Executive Order N-9-26 served from the Governor's website, and its text extraction, read end to end from the recitals through the numbered orders, the filing and no-rights paragraphs and the attestation dated the 18th day of September 2026. The signature lines render as blanks in the text extraction; on the rendered page 3 the Governor's signature is visible above his printed name, and the Secretary of State's ATTEST line is blank, so we do not say the order was attested or filed. The retrieved bytes matched the live official PDF at the time of our gate check. We also read the chaptered texts of SB 813 and AB 1405 as retained from the Legislature's bill-text pages, for the section numbers, the statutory dates and the subdivisions that the order's paragraphs 1 and 2 reference; we did not read either bill end to end.

What we did not open: the Governor's press release of 18 September 2026, which returned an access error to our fetches, so nothing in this article rests on it; Executive Orders N-12-23 and N-5-26 and the June 2025 frontier AI report that the recitals mention; SB 53's operative text; any Secretary of State filing record; and any agency work product on the November recommendations. We describe those instruments only as this order describes them.

What we refuse to claim: we do not say California now requires a kill switch, onsite auditors or independently verified safety frameworks, because paragraph 3 asks for recommendations on the feasibility and potential efficacy of such requirements and imposes none. We do not say that SB 53, SB 813 or AB 1405 will be amended, because the order names no bill in paragraph 3 and recommendations are not legislation. We do not say the order moves the January 1, 2029 registration prohibition in section 11549.82.5, because paragraph 2 references section 11549.82 only. We do not say subdivision (b) of section 11549.82 has to be completed by December 1, 2027, because the order says begin. We do not treat the recitals' descriptions of AI incidents as established fact. We do not say the order has been filed with the Secretary of State. Where the word must or shall appears above, it is quoting or closely reporting the order or the enacted statutes.

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.

Related briefings

• California SB 813 and independent verification organizations
• California SB 53 frontier AI transparency duties in force
• California AI workforce executive order and WARN Act recommendations
• North America AI regulation tracker
• AI Regulation News hub