California Generative AI Transparency Requirements - AB 2013
Issued by
California State Legislature
- September 30, 2026 · Correction — Added the official link to the bill text. (Cody Maxwell)
- September 30, 2026 · Correction — Replaced an invented bias and limitations requirement with the statute's actual disclosure items, and added its 2022 release-date threshold (fact-check finding). (Cody Maxwell)
- October 1, 2026 · Correction — Rewrote the practical steps, newsletter hook, search description, and audience fields to match the entry's corrected content. (Cody Maxwell)
California Assembly Bill (AB) 2013 requires public training-data disclosures from developers of generative AI systems accessible to the California public. Developers must publish the documentation on their websites. The disclosures explain training datasets’ origins and composition. The requirement applies regardless of where the developer is headquartered.
Applies To
Overview
AB 2013 took effect January 1, 2026, and establishes a transparency baseline for generative AI training data in California. Covered developers must post publicly accessible documentation on their websites describing the datasets used to train their generative AI systems. The documentation is a high-level summary of the training data, covering items such as sources, data types, copyright status, personal information, and processing. It applies to generative AI systems released on or after 1 January 2022 and made publicly available to Californians. Enforcement authority and specific penalty provisions are administered under California's existing consumer protection and business regulation framework. The legislation does not establish a dedicated AI regulator but relies on existing state enforcement mechanisms. AB 2013 is part of a broader package of California AI legislation enacted in 2024 aimed at increasing accountability across the generative AI supply chain.
Key Requirements
- •Developers must post a high-level summary of training data on their website, with compliance due by 1 January 2026.
- •The summary covers dataset sources or owners, purpose, size, data types, and copyright or public domain status.
- •It also covers whether data was purchased or licensed, includes personal or aggregate consumer information, was cleaned or modified, and the collection period.
- •It also states when the data was first used and whether synthetic data was used.
- •It applies to systems released on or after 1 January 2022 and made publicly available to Californians.
- •Systems for security, aircraft operation, or federal national security use are exempt.
What Your Organization Must Do
- →Inventory generative AI systems your organization released on or after 1 January 2022 and made available to Californians.
- →Draft a high-level training data summary covering sources, size, data types, and copyright or public domain status.
- →Record whether training data was purchased or licensed, and whether it included personal or aggregate consumer information.
- →Document any cleaning or modification of datasets, collection periods, first use dates, and use of synthetic data.
- →Publish the documentation on your public website, since posting on the site is the stated compliance route.
- →Check whether a system falls within the exemptions for security, aircraft operation, or federal national security use.