California AI Auditor Registry and Evaluator Certification Law Guide: AB 1405 Duties, Dates and Audit Readiness

From 1 January 2029 nobody may offer, sell or conduct an audit of an AI system's compliance with California law unless they are on a state AI Auditor Registry, and the state is also building a separate designation for independent verification organisations that assess AI risk. Governor Newsom signed both laws on 9 September 2026: AB 1405 (Chapter 178, Statutes of 2026) creates the AI Auditor Registry, and SB 813 (Chapter 179, Statutes of 2026) creates the independent verification organisation (IVO) framework. Both are run by the Government Operations Agency (GovOps). Neither law forces an AI developer to be audited, but other California laws increasingly call for audits and assessments, and when they do, who may perform them and what the report must contain will be set here. This guide is for AI governance leads, internal audit, assurance firms and security teams who commission or deliver AI audits, and reflects the chaptered bills as at 11 November 2026. For California's wider AI rulebook see the California AI Laws Guide.

Source note: provisions are taken from the chaptered bill pages on CalMatters Digital Democracy, checked against an Akin Gump analysis; the official leginfo.legislature.ca.gov pages could not be read. Both laws are enacted; their main obligations start in 2028 and 2029. Confirm against the official text before relying on it. Compliance steps, decision guidance and checklists on this page are NHI Mgmt Group recommendations, not requirements of the text.

Key takeaways

• Two laws, two roles. AB 1405 registers anyone who conducts a covered AI audit; SB 813 designates a smaller set of IVOs with demonstrated expertise in assessing AI risk.
• Dates are set in statute. GovOps must finalise the IVO application requirements and criteria by 1 January 2028, and stand up the public AI Auditor Registry by 1 January 2029, when unregistered audits become prohibited.
• Audits are not mandated by these laws. SB 813 states it does not require developers to use an IVO or undergo a covered audit; the obligations fall on auditors and on GovOps.
• Independence rules are concrete. No auditing your own work, no team members who held a material role at the auditee in the prior 12 months, no fees conditioned on results, and records kept for at least 10 years.
• Audit evidence carries weight in court. Under SB 813, an audit that followed an identified standard is relevant to, but not conclusive of, a lawsuit alleging harm from an AI system.

Scope: who and what is covered

Key definitions

• Artificial intelligence: an engineered or machine-based system with varying levels of autonomy that infers from its inputs how to generate outputs that can influence physical or virtual environments. This matches the definition used across recent California AI statutes.
• AI auditor: a person, partnership or corporation (SB 813 adds academic institutions and nonprofits) that assesses an AI system or model, or conducts a covered audit, on behalf of a third party. In-house teams auditing their own employer's systems are not acting for a third party.
• Covered AI audit: an audit of the internal controls, processes or systems that an AI system or model uses, or needs, to comply with state law. A general model evaluation or red team exercise that does not assess state-law compliance sits outside this definition.
• Independent verification organisation: an AI auditor that GovOps designates as having demonstrated expertise in assessing AI risks and in identifying the metrics and methodologies for that assessment.

Key obligations for registered AI auditors (AB 1405)

Registration

• Information to file: business name and contact details; the California laws or regulations under which the auditor conducts covered audits; relevant certifications or accreditations; a written service description of no more than 500 words; and a standard operating procedure naming the standards followed (the statute gives ISO, NIST and accountancy bodies as examples) and the basis for any accuracy or reliability claims. Trade secrets may be redacted.
• Updates: material changes must be filed within 90 days.
• Fees: annual fees set by GovOps, capped at the reasonable cost of administration, paid into the AI Auditors Registration Fund.
• Advertising: the registration number must be displayed prominently in advertising for covered audit services. The registry carries a notice that registration is not state endorsement.

Audit conduct and the audit report

Audits must follow widely recognised industry standards suited to the system, where they exist. Every report must contain the elements below, which makes the law usable as a report template today.

Audit information and supporting documentation must be kept for at least 10 years. That retention period outlasts most model lifecycles, so evaluation artefacts, prompts, outputs and logs need an archival plan from the first engagement.

Independence, objectivity and integrity

• Decline any audit where a financial, business, employment or other relationship would reasonably impair independence or objectivity; reasonable audit fees alone are not a prohibited interest.
• Do not evaluate your own work, or a system you designed, built, implemented or operated for the auditee.
• Audit staff may not seek or accept employment with the auditee during the engagement.
• Do not assign anyone who held a material role in the audited subject matter at the auditee in the previous 12 months.
• Exercise impartial professional judgement and ensure the team has the competence the system demands.
• Do not stop employees reporting suspected non-compliance to the Attorney General or Labor Commissioner, or filing a misconduct report, and do not retaliate against them.

Licensed California CPAs and firms with a valid permit to practise public accountancy are treated as meeting the report and independence requirements if they comply with the California Accountancy Act and AICPA standards.

The IVO framework (SB 813)

SB 813 asks GovOps to define what a credible AI risk assessor looks like. By 1 January 2028 GovOps must publish application requirements, designation criteria, and procedures to suspend or terminate a designation. At a minimum an IVO must:

• Assess AI risk with stated methods: identify the benchmarks, technologies, metrics and methodologies it will use, and submit them with its application.
• Have the people: employ or engage personnel with sufficient technical expertise.
• Manage conflicts: it may be paid at reasonable market rates, but not with payment conditioned on assessment results.
• Stay independent: no operational or management dependence on the assessed party.

Grounds for suspension or termination include failing to follow standards, material misrepresentation, conflicts that impair independence, inadequate documentation, conduct that calls integrity or competence into question, and cybersecurity lapses. GovOps must align criteria with existing audit and assurance standards where practicable, and minimise duplicate compliance, including by accepting qualifying reports prepared for substantially similar requirements. It must consult widely and convene working groups that include engineers from competing AI companies and AI safety experts, then report findings to the Legislature.

Designated IVOs report annually to GovOps and the Legislature, starting no sooner than 12 months after designation, on their standards and methodologies and on governance or funding changes relevant to independence. Redactions are allowed for trade secrets, cybersecurity, public safety and national security, but must be described, with unredacted material kept for five years.

Timeline and current status

GovOps may adopt regulations, so detail on fees, forms and professional standards is still to come. Funding for the registry depends on legislative appropriation. Treat timelines as statutory deadlines that agencies sometimes miss, and check GovOps announcements before planning around them. The registry chapter also states that it does not affect audits required under other laws that take effect before it does, so an audit regime that starts earlier keeps its own rules on who may audit until the registry applies.

Enforcement and penalties

• Investigation: GovOps may investigate after a misconduct report and request information needed to decide whether a violation occurred. Misconduct reports are kept for the registration period plus 10 years and may be shared with other state enforcement agencies.
• Sanctions: a violation is grounds for removal from the registry and referral to the Attorney General or another enforcement authority, after written notice and a reasonable opportunity to cure or contest. No fine schedule appears in the chaptered summaries reviewed.
• CPAs: findings against a licensed CPA or firm are sent to the California Board of Accountancy, which investigates under its own procedures.
• Litigation weight: SB 813 creates no safe harbour or presumption, and no liability solely for failing to meet a standard; audit evidence is relevant but not conclusive.

Decision guidance

• AI developers and deployers: you are not obliged to commission an audit under these laws. Where another law or a contract requires one that assesses California-law compliance, from 2029 hire a registered auditor and check its registration number. The 2026 companion chatbot law SB 1119, for example, requires independent child safety audits for chatbot operators.
• Assurance firms and consultancies: decide by 2028 whether you will register, and whether you will seek IVO designation. Separate advisory and audit work for the same client now, because the self-review and 12-month cooling-off rules will bite on existing relationships.
• Internal audit: internal teams auditing their own organisation are not acting for a third party, but adopting the report elements now gives a consistent evidence standard and eases a later external audit.
• Small vendors: rely on GovOps' commitment to accept substantially similar reports; an ISO/IEC 42001 certification audit or SOC-style report may reduce duplicate work, subject to the final criteria.

What it means for AI agents and non-human identities

An auditor of an agentic system will need evidence of what the agent was allowed to do and what it did. That means an inventory of agent identities and their credentials, scoped permissions, and tamper-evident logs of tool calls and decisions, retained long enough for a 10-year audit file. Auditor access is itself a risk: grant read-only, time-bound accounts or evidence exports rather than standing credentials to production. See the AI Agent Audit Trails for Regulators Guide and Audit Readiness for AI Agents and NHIs.

Compliance steps

1. Map audit triggers. List every California law, contract or customer requirement that calls for an audit or assessment of your AI systems, and mark which assess state-law compliance.
2. Adopt the report template. Use the six AB 1405 report elements in internal reviews now, and pilot them on one system before rolling out.
3. Build the evidence base. Inventory models and agents, version evaluations, and keep logs and test artefacts under a 10-year retention rule.
4. Screen auditors. Before 2029 ask prospective auditors for their standards SOP, independence checks and team histories; from 2029 verify registration.
5. Handle access safely. Provide auditor accounts that are least privilege, time-bound and logged; revoke at engagement close.
6. Watch GovOps. Track regulations, fee schedules and the IVO criteria due by 1 January 2028; adjust contracts and templates when they land.

Framework mapping and measurement

Metrics: percentage of AI systems with a current audit or assessment; open audit findings past due date; percentage of external AI auditors with verified registration (from 2029) and documented independence checks; evidence retrieval time for a sampled decision; and auditor accounts still active after engagement close (target zero).

Common failure patterns

• Assuming the law requires your audit. It regulates auditors; the duty to be audited comes from other laws or contracts.
• Using the implementer as auditor. A firm that built or operated the system cannot audit it.
• Short retention. Deleting evaluation data after a model is retired leaves auditors unable to meet the 10-year record rule.
• Reports with no limitations section. Overstated assurance is a misconduct and litigation risk.
• Counting a model evaluation as a covered audit. A benchmark run or red team exercise that does not assess compliance with a California law is not a covered AI audit; label it accurately so registration questions do not arise later, and so customers are not told they hold an audit they do not have.
• Standing auditor credentials. Accounts left active after fieldwork become unmanaged non-human or third-party access.

Practitioner checklist

• Record Chapters 178 and 179, Statutes of 2026, and the 2028 and 2029 deadlines in the regulatory register.
• Identify every audit or assessment trigger for AI systems in California.
• Adopt the six-element audit report template internally.
• Set 10-year retention for AI audit evidence.
• Add independence and 12-month cooling-off checks to auditor selection.
• Provision auditor access as least privilege, time-bound and logged.
• Assign an owner to track GovOps regulations and IVO criteria.
• From 1 January 2029, verify the registration number of any AI auditor.

Standards and references

• CalMatters Digital Democracy: AB 1405 Artificial Intelligence Auditors Registration, chaptered bill
• CalMatters Digital Democracy: SB 813 Independent Verification Organizations, chaptered bill
• Akin: California Enacts AI Auditor Registry and Independent Verification Framework
• WilmerHale: California Finalizes Next Wave of State AI and Privacy Regulation
• ISO: ISO/IEC 42001:2023 Artificial intelligence management system
• NIST: AI Risk Management Framework
• NIST: SP 800-53 Rev. 5 Security and Privacy Controls

This guide summarises California AB 1405 and SB 813 as at 11 November 2026, and is not legal advice. Related NHI Mgmt Group resources: California AI Laws Guide · ISO/IEC 42001 Guide · Third-Party AI Risk Guide · AI Agent Audit Trails for Regulators Guide · Audit Readiness for AI Agents and NHIs