California Legislature Passes Independent AI Safety Verification Bill

The California Legislature gave final approval on August 30, 2026, to SB 813, legislation authored by Senator Jerry McNerney that would direct the Government Operations Agency to designate independent verification organizations — outside auditors qualified to assess the safety risks of AI systems and models. The bill now goes to Governor Gavin Newsom.

The Assembly passed the bill on third reading on August 30, 2026, and the Senate concurred in the Assembly’s amendments the same day, completing legislative passage, according to the official vote record. The Assembly recorded 67 ayes, 6 noes, and 6 members not voting; the Senate concurrence passed with 39 ayes, no opposition, and one member not voting.

What SB 813 Would Require

The bill adds a new chapter to the California Government Code covering what it titles California Artificial Intelligence Safety Independent Verification Organizations. On or before January 1, 2028, the Government Operations Agency must develop application requirements for designation as an independent verification organization, or IVO; procedures for suspending or terminating a designation; and criteria for determining whether an AI auditor qualifies. The agency must publish those requirements and criteria publicly, review and revise them as technology and industry standards change, and, where practicable, align them with existing professional and regulatory audit and assurance standards.

Under the bill’s definition, an IVO is an AI auditor that has demonstrated expertise in assessing the risks posed by an AI system or model and in identifying the metrics and methodologies behind that assessment. A designated IVO must have no financial, operational, or management dependence on the party it assesses. It may accept payment from an assessed party at reasonable market rates, but it may not accept terms that condition payment on the results of its assessment.

The suspension and termination procedures must cover failures to adhere to appropriate standards, material misrepresentations in applications or required disclosures, conflicts of interest that impair independence, failures to maintain adequate documentation, lapses in cybersecurity, and conduct that reasonably calls into question an IVO’s integrity, objectivity, or competence.

The agency must also convene working groups that include engineers from competing AI companies and AI safety experts to gather stakeholder input on standards, and must report the working groups’ findings to the Legislature. Each designated IVO must submit an annual report, no sooner than 12 months after its initial designation, summarizing its standards and methodologies and disclosing changes to governance policies or funding sources relevant to its independence. IVOs may redact material to protect trade secrets, cybersecurity, public safety, or national security, but must retain unredacted information for five years.

The framework is voluntary. The bill states that it does not require any developer, deployer, or operator of an AI system or model to engage an IVO or undergo an audit, does not establish liability solely for failure to comply with a standard, and does not constitute a state recommendation or endorsement of any AI system or model. In litigation alleging that an AI system caused harm, the bill provides that an audit performed in accordance with a standard identified under the chapter is relevant to, but not conclusive of, the action.

Author and Sponsor Statements

McNerney’s office described the measure as a first-in-the-nation framework for independent, third-party assessments of potential safety risks posed by AI systems. His announcement reported the Senate vote as 37–0 and the Assembly vote as 53–4; the official legislative record lists the Senate concurrence at 39–0 with one member not voting and the Assembly vote at 67–6 with six members not voting.

“The California Legislature today sent a clear message: It’s time for California to take the lead in assessing AI’s potential safety risks, since Washington, D.C., is unable or unwilling to do so,” McNerney said in a statement announcing the vote. His office also said that establishing independent, third-party evaluations of AI was one of the primary recommendations of Newsom’s blue-ribbon panel on artificial intelligence, and cited polling indicating that two-thirds of Americans have little or no confidence in the federal government to regulate AI effectively.

SB 813 is sponsored by Fathom, a nonprofit that says it developed and championed the IVO model and works with policymakers across the country to put it into practice. The organization described IVOs as qualified outside experts that can independently assess the risk of AI models and systems using credible testing and analysis, and said the bill would let the potential risks posed by the technology be independently assessed rather than taken on faith.

“The power of the independent verification model is that it lets independent, qualified experts assess the risk of AI capabilities, and it keeps pace with the technology instead of freezing a single test into law,” said Andrew Freedman, Fathom’s CEO, in the organization’s statement.

Fathom’s statement also placed the bill alongside SB 53, which it said now requires large frontier developers to publish a safety framework and report critical incidents. The organization said SB 813 would begin building a further layer of accountability because safety determinations still largely rely on companies assessing their own systems, and that it will continue working with California policymakers, technical experts, and industry stakeholders to move the framework forward.

Earlier Versions of the Measure

McNerney introduced SB 813 on February 21, 2025, with Assemblymembers Bauer-Kahan and Lowenthal as coauthors, and the bill was amended repeatedly in both chambers, with Assembly amendments as recent as August 26, 2026, according to the legislative record. The Senate passed an earlier version of the bill 31–7 on January 27, 2026. McNerney’s announcement at the time said that version would have created a California Artificial Intelligence Standards and Safety Commission to designate IVO expert panels that would set voluntary standards, certify AI developers and vendors, and monitor compliance, with the power to revoke certification.

If Newsom signs the bill, the Government Operations Agency faces a January 1, 2028 deadline to build the designation system, with working-group findings to be reported to the Legislature and the first annual IVO reports due no earlier than 12 months after an organization’s initial designation.