Governor Newsom signed AB-2013 into law, making California the first US state to require training-data disclosures for generative AI products sold or offered to California customers. The law takes effect January 2027. Providers must publish a public summary covering data sources, modalities, scale, and copyright posture.
The text reads as lighter-touch than the EU AI Act’s parallel obligation, but the implementing regulations from the California Privacy Protection Agency will set the binding detail. The draft template circulated during legislative markup asked for source-category breakdowns at a granularity comparable to the EU template, including identification of major commercial datasets and proprietary corpora. If that survives into the final rules, US providers will have a single disclosure surface that satisfies both EU and California, which is the practical outcome a lot of compliance teams have been quietly hoping for.
The political read: federal preemption arguments are now significantly weaker. Other states (NY, WA, IL) have similar bills in committee and will move faster with California as cover. Expect a coordinated federal-vs-state fight inside the next twelve months.