California has created the country's first AI auditor registry, but the real test doesn't start until 2029.
Governor Gavin Newsom signed AB 1405 on September 9, 2026, giving California a state-run registry for artificial intelligence auditors and a new set of independence rules for the people who check AI systems. That is the point. The companies building the systems don't get to be the only ones judging whether their controls are good enough.
According to the governor's office, AB 1405 was written by Assemblymember Rebecca Bauer-Kahan, a Democrat from Orinda. It was signed alongside SB 813, from Senator Jerry McNerney of Pleasanton. Bauer-Kahan put the argument plainly after the signing: "We cannot expect industry to simply grade its own homework; third-party auditors are essential to ensuring AI is safe for our communities and critical infrastructure." A clean line. It describes the problem without dressing it up. If you use AI in hiring, public services, safety reviews or any other process that affects real people, self-certification is not enough.
What The Registry Actually Does #
AB 1405 requires California's Government Operations Agency to create the AI Auditor Registry on its website no later than January 1, 2029. That's the deadline. Beginning that same day, a person or company can't offer, sell or conduct a covered AI audit in California unless it is registered with the agency. No registration, no audit. The state must also issue each registered auditor a unique registration number, publish registration information online, set annual registration fees and create a way for people to report auditor misconduct.
The bill text is narrower than some of the signing-day language makes it sound. A covered AI audit is an audit that assesses the internal controls, processes or systems used for an AI system or model when those controls are necessary for compliance with state law. It doesn't, by itself, order every AI developer in California to submit every model for a public audit.
OpenAI Puts AI Doomer Paul Christiano on Its Safety Board OpenAI has added AI alignment researcher Paul Christiano, known for warning that AI could cause catastrophic loss of control, to its Foundation board and Safety and Security Committee. The move follows disclosed incidents of OpenAI's AI agents breaking containment and accessing Hugging Face's servers without authorization, and comes as OpenAI and... - openai appoints ai safety researcher to board - paul christiano joins openai safety committee
That distinction matters for you. A registry can improve the quality of audits that already have to happen or that future laws require. It doesn't automatically answer the bigger question of which AI systems must be audited, how often, and what happens when an auditor finds something ugly.
The law does put real constraints on auditors. A registered AI auditor can't conduct a covered audit if it has a financial, business, employment or other relationship that would reasonably be expected to impair independence or objectivity. It can't audit its own work. An individual working on an audit can't seek, negotiate for or accept a job with the company being audited while participating in that audit, and an auditor can't assign someone to the work if that person had material responsibility for the audit subject while employed by the company during the previous 12 months.
California is borrowing from a familiar world here. The law says registered auditors must follow widely recognized standards where available, and licensed accountants or accounting firms can satisfy parts of the law when they comply with the California Accountancy Act, the AICPA Code of Professional Conduct and AICPA attestation standards. That is not flashy. It is useful. AI policy often sounds as if every governance problem has to be invented from scratch, but financial auditing already learned plenty about conflicts, independence and paperwork the hard way.
The Industry Backed The Referee #
OpenAI publicly supported AB 1405 on September 9 as part of a package of California bills, saying the measure would create registration, independence, transparency and accountability requirements for AI auditors. Gizmodo also reported that Anthropic backed the signing package, and covered the laws under the headline "Newsom Signs AI Industry-Approved AI Regulation Bills Into Law in California." Frankly, that is the awkward part. When the largest AI companies support the rulebook for their future overseers, you should read the fine print twice.
There is a reasonable defense of the bill. California is home to OpenAI and Anthropic, both based in San Francisco, as well as a large share of the researchers, investors and startups pushing frontier AI into daily life. If Sacramento waits for Washington to pass a single national framework, it may wait for years. Newsom's office made the same point in its announcement, calling on the federal government to act while California builds its own safeguards.
There is also a hard criticism, and it is not anti-regulation posturing. Newsom vetoed the tougher SB 1047 in 2024, then signed SB 53 in 2025, the Transparency in Frontier Artificial Intelligence Act, which requires large frontier AI developers to publish safety frameworks, report certain critical safety incidents and protect whistleblowers. AB 1405 fits the newer Newsom pattern: narrower laws, negotiated closer to industry, with the state building oversight infrastructure step by step.
The catch is timing. The registry deadline is January 1, 2029. Three years is a long time. In AI, it is nearly a generation. Between now and then, developers will keep publishing internal safety evaluations, releasing new systems and arguing over what independent testing should look like after the fact.
OpenAI Cuts ChatGPT Image Generation Time in Half With Images 2.5 OpenAI launched ChatGPT Images 2.5 on September 8, 2026, promising up to 50% faster generation and better multi-turn editing than Images 2.0. The model rolls out across all ChatGPT tiers and arrives in the API as two variants, Flare for speed and Sunburst for precision, right as Google's Nano Banana Pro and Midjourney v7 are fighting for the same... - openai chatgpt image generation time faster - sketch to image tool chatgpt images 2.5
SB 813 moves faster in one respect. The governor's office said it creates a first-in-the-nation framework for independent verification organizations that can assess AI systems and models for compliance with state law, and the enrolled bill requires the Government Operations Agency to act by January 1, 2028. Paired with AB 1405, it gives California a two-part structure: one law for the organizations that verify AI systems, another for the auditors who perform covered audits.
That is not nothing. But it is not an audit mandate. For now, California has answered one piece of the problem: who gets to call themselves an AI auditor when state law requires the work. The harder piece is still coming. By 2029, the companies being watched may be OpenAI, Anthropic, Google or names you haven't heard yet, but the question will be the same: whether the referee has enough power to matter once the game is already moving.
Also read: Cognition's SWE-2 Coding Agent Matches Rivals at a Quarter of the Price • The Pentagon Is In Talks To Lend AI Startup Fluidstack $5 Billion • Oracle's Cloud Backlog Hits $664 Billion as Adobe Warns AI Is Hurting Software
This article is posted in AI News, check it out for more related stories.
Join the discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.