Summary
On September 9, 2026, California Governor Gavin Newsom signed two bills that regulate the people who audit AI systems rather than the systems themselves. SB 813, by Senator Jerry McNerney, directs the Government Operations Agency to designate "independent verification organizations" with demonstrated expertise in assessing AI risk by January 1, 2028; AB 1405, by Assemblymember Rebecca Bauer-Kahan, requires the same agency to open an AI Auditor Registry by January 1, 2029, after which no unregistered person may offer or conduct a covered AI audit in the state. Neither law requires any developer or deployer to undergo an audit. OpenAI published a global-affairs post the same day formally endorsing both bills, along with two others, and stating that some of them it "did not endorse in the past."
What Happened
The two bills reached the governor's desk in the final days of the 2025-26 session. AB 1405 was introduced in February 2025, passed the Assembly 62-4 in June 2025, then sat in the Senate Appropriations suspense file from August 2025 until August 13, 2026, when it was revived; the Senate passed it 29-10 and the Assembly concurred in the Senate amendments 60-6 on August 30. SB 813 passed the Senate 31-7 in January 2026, was rewritten in the Assembly over the summer, passed the Assembly 67-6 and cleared Senate concurrence 37-0 on August 30. Both were chaptered by the Secretary of State on September 9 as Chapters 178 and 179 of the Statutes of 2026.
SB 813 adds a new chapter to the Government Code establishing "independent verification organizations," defined as AI auditors that the Government Operations Agency has designated as having demonstrated expertise in "assessing the risks posed by an AI system or model and identifying the metrics and methodologies that form the basis for that assessment." By January 1, 2028, the agency must publish application requirements and designation criteria, drawing on standards from government agencies, national and international auditing bodies, and the industry, and must set procedures for suspending or terminating a designation for causes including material misrepresentation, conflicts of interest and "lapses in cybersecurity." The criteria must at minimum cover risk-assessment competence, technical staffing, and management of conflicts of interest "including ensuring financial relationships do not impact the judgment and decisionmaking of the IVO." The agency must convene working groups that include "engineers from AI companies that are competitors and AI safety experts" and report their findings to the Legislature. Designated organizations must file an annual report summarizing their standards and methodologies and any changes to governance or funding relevant to independence, with redactions permitted for trade secrets, cybersecurity, public safety and national security, and unredacted copies retained for five years. A limitations section states that the chapter does not "establish liability solely for failure to comply with a standard," does not constitute state endorsement of any AI system, and does not require anyone who develops, deploys or operates an AI system to engage an IVO or undergo an audit; in litigation alleging that an AI system caused harm, an audit performed under a recognized standard "is relevant to, but not conclusive of, the action."
AB 1405 adds a separate chapter requiring the agency to establish a public AI Auditor Registry, fix annual registration fees capped at the cost of administration, and provide a mechanism for members of the public to report auditor misconduct, all by January 1, 2029. From that date, an unregistered person may not offer, sell or conduct a "covered AI audit," defined as an audit assessing the internal controls, processes or systems implemented for an AI system or model that are necessary for compliance with state law. Registrants must supply a business name and contact details, the California laws under which they audit, any certifications, a description of their services of no more than 500 words, and a standard operating procedure referencing the standards they apply, with ISO, NIST and national and international auditing and assurance bodies named as examples. Each registered auditor receives a registration number that must appear on all advertising. The law imposes independence rules modeled on accounting practice: an auditor may not seek or accept employment with an auditee during an audit, may not audit where a financial, business or employment relationship would reasonably be expected to impair independence, must give the auditee a signed and dated statement that the audit followed the statute, and may not retaliate against employees who engage in whistleblower activity. Violations are grounds for removal from the registry and referral to the Attorney General. Certified public accountants and accounting firms in good standing with the California Board of Accountancy are exempt from some reporting and independence provisions, with complaints against them routed to the Board. Fees go to a new AI Auditors' Registration Fund, available for expenditure only on appropriation by the Legislature.
The governor's office framed the pair as "first-in-the-nation standards for third-party audits and independent assessments of AI systems" and placed them in a lineage running from the 2023 executive order, the 2024 package on deepfakes and watermarking, and SB 53 in 2025. Newsom's statement said "the federal government must step forward with robust, national regulations that match the urgency of this moment." McNerney said the law "codifies one of the primary recommendations of the governor's blue-ribbon panel on AI" and referred to reports "just this week" that "the most powerful AI systems teamed with AI agents pose real threats to humanity." Bauer-Kahan said the state "cannot expect industry to simply grade its own homework."
OpenAI's post, signed by Chief Global Affairs Officer Chris Lehane and dated the same day, announced that the company was "formally endorsing four additional California bills that have passed the legislature": SB 813, AB 1405, SB 1119 on companion chatbots and minors, and AB 1864 on screening standards for gene-synthesis providers. On SB 813 the post said OpenAI would "prefer independent technical assessments to be required at the federal level" but that "in the absence of federal action" California "can help establish the rules of the road for a secure, capable national independent-assessment system." It added: "Some of these bills we did not endorse in the past, and are now supporting after reconsidering in light of the recent jump in capabilities we have seen." The same post called for "mandatory, capability-based national AI safety regulation," listed SB 53, New York's RAISE Act, Illinois's SB 315 and a Massachusetts proposal as state laws the company had supported, and described the strategy of converging state rules into a de facto national baseline as "reverse federalism." OpenAI had opposed SB 53 before it passed and in August 2026 asked California to amend it to require monitoring of frontier models under training or evaluation, according to TechCrunch.
The new laws create a supply of accredited auditors but no statutory demand for their services. SB 53, which took effect January 1, 2026, requires frontier developers to publish safety frameworks and report incidents but contains no audit mandate. Illinois's SB 315, passed by that state's legislature in May 2026, would require annual independent third-party audits of frontier developers, which NBC News reported would be a first in US AI legislation.
Why It Matters
Many AI-governance proposals since 2023 have assumed that someone will audit the systems; few have said who, under what standard, or with what independence. These two statutes are the first US attempt to answer the "who" question in law. By licensing auditors before mandating audits, California is building the institution it would need if SB 53 or a successor were later amended to require third-party assessment, and it is doing so in the state where most frontier developers are headquartered. The "relevant to, but not conclusive of" clause in SB 813 is the provision most likely to matter in practice: it gives an audit under a recognized standard some evidentiary weight in a harm suit without creating a safe harbor, so the value of a designation will be set as much by courts as by the agency.
OpenAI's endorsement is a data point on how the politics of state AI regulation shifted over the summer of 2026. A company that lobbied against SB 53 in 2025 now writes that it "did not endorse" some of these bills in the past and changed position because of "the recent jump in capabilities," language that follows its own disclosure of the Hugging Face incident, its chief scientist's call for voluntary slowdowns, and the Senate subcommittee investigation opened the following day. Whether that support survives a federal preemption push, such as the three-year moratorium in the House discussion draft, is not known.
Several things remain open. The Government Operations Agency has sixteen months to write IVO criteria and twenty-eight to open a registry, with no appropriation attached to either bill beyond the fee-funded registry account. The laws say nothing about who pays for verification or what access to weights, training data or internal evaluations an IVO may demand, questions that OpenAI's post explicitly says it would rather see settled federally. Whether any frontier developer voluntarily submits to a designated IVO before a statute requires it, and whether the registry standards converge with or diverge from the Illinois audit mandate and the EU's conformity-assessment regime, will determine whether this is the start of an audit profession or an unused registry.
§ How to read the metadata
- Landmark
- Fundamentally alters the trajectory; 2–5 per year.
- Major
- Meaningfully shifts the landscape; 2–4 per month.
- Notable
- Worth documenting; significance can be upgraded later.
- Confidence
- High = primary sources corroborate. Medium = credible secondary only. Low = provisional. Disputed = credible sources disagree.
- Contestation
- Uncontested = no formal challenge. Contested = at least one challenge open. Superseded = replaced by a later entry. Unresolved = dispute still open.