Gov. Gavin Newsom signed SB 813 and AB 1405 on September 9, standing up the first state-level independent AI audit framework in the country and, more consequentially, extending its reach to any company deploying AI in a high-stakes decision. Not just the labs. The deployers.

SB 813, authored by Sen. McNerney, creates the California Artificial Intelligence Standards and Safety Commission, which will recognize Independent Verification Organizations. AB 1405, from Assemblymember Bauer-Kahan, establishes the state auditor registry itself and sets standards for independence, transparency, and integrity. SB 813 cleared the Senate 37-0 and the Assembly 53-4, a margin that tells you how the political vibes shifted after California’s 24-bill AI package sent to Newsom last week.

The Government Operations Agency must certify the first IVOs by January 1, 2028. Methodology standards don’t yet exist. Per TechTimes, the framework’s final quality depends entirely on what methodology standards the Commission produces before the January 2028 IVO deadline, with academic researchers warning of an “audit-washing” failure mode in which certified auditors provide the appearance of accountability without substantive verification.

Both Anthropic and OpenAI backed the bills, with OpenAI reversing its earlier opposition, a reversal that reads less like conversion than an acknowledgment that federally preempted AI governance isn’t arriving in time.

The framework’s real bite is scope. Its threshold is deployment in a high-stakes context, not model training, which means any business using an off-the-shelf model to screen job applicants, price insurance, or make consequential customer calls is inside the perimeter. It layers on top of California’s automated decision-making rules that took effect January 1, 2026, covering housing, insurance, healthcare, employment, and essential services.

A separate bill, SB 1119, or “Adam’s Law,” requires companion-chatbot operators to run risk assessments and independent child-safety audits. Combined with the AI ad-disclosure clock and Texas TRAIGA’s now-open complaint portal covering AI sales tools, the state-level enforcement lattice is now legible enough to plan against.

Sources

Sources