California just did something no other state has done: it created a system to license the people who check AI systems for safety and fairness.
On September 9, Governor Gavin Newsom signed two bills, Senate Bill 813 and Assembly Bill 1405. Together they build the first government backed system in the country for training, vetting, and registering outside auditors who inspect AI tools on behalf of businesses and regulators.
Think of it like the accounting profession. Companies do not just tell investors their books are fine. They pay a licensed accountant to check the books and put their name on it. California is trying to build the same setup for AI: instead of a company just saying "our hiring tool is not biased" or "our chatbot is safe," a licensed, registered outside auditor checks the claim and signs off.
Here is how the two laws split the work. Senate Bill 813 sets up a new state commission that will approve specific organizations, called independent verification organizations, as qualified to test AI models for safety risks. Assembly Bill 1405 goes further down the chain: it creates an actual public registry of individual AI auditors, with rules about who counts as independent and who does not. Starting in 2029, anyone doing this kind of audit in California without being on that registry will be breaking the law.
Nobody needs to scramble this week. The registry does not open until 2029, and the verification organization framework has until 2028 to get built out. This is a groundwork law, not a compliance deadline.
But the direction matters more than the date. A market for checking AI systems already exists today on a voluntary basis. Firms like BABL AI, ORCAA, and Holistic AI already audit hiring algorithms and other automated tools for companies that want to get ahead of laws like New York City's bias audit rule or the EU's AI Act. California is not creating this industry from nothing. It is putting a state seal of approval on it, which is exactly what tends to happen right before a voluntary practice turns into a mandatory one.
There is no federal AI law covering any of this, and there is unlikely to be one soon. A Senate attempt to block states from regulating AI at all was stripped from a broader bill on a 99 to 1 vote last year. With Washington stuck, California's rules tend to become the default for companies that sell into a state of 39 million people and do not want to build a separate version of their product for everywhere else.
Even OpenAI, which usually prefers one federal rulebook over fifty state ones, backed these bills. That is telling. The company would rather have a credible, licensed audit system exist somewhere than have none at all, because a system like this is what lets an AI company say "we were checked by someone independent" instead of just "trust us."
For any business that uses AI to screen job applicants, price insurance, approve loans, or make other decisions that affect people, this is the early shape of what "proving your AI is safe" will look like a few years from now: not a marketing claim, but a credential from someone licensed to check.