California AB 2013
The Training Data Transparency Act took effect on 1 January 2026. It asks generative AI developers to post a summary of what they trained on. No template, no filing, no central register — so nobody has been able to say who actually did it.
1 disclosure found so far · last swept 16 August 2026
What we have found
On 16 August 2026 we probed the legal, transparency, policy and training-data pages of fifteen major model developers — 210 URLs — for a document identifiable as an AB 2013 disclosure: one citing the Act, the Training Data Transparency Act, or Civil Code § 3111.
We found one.
Read that as a statement about our search, not a verdict on anyone. AB 2013 prescribes no format and no location, so a compliant disclosure could sit inside a terms page, a dataset appendix or a PDF we did not think to open. We are looking for a document that names the law; a developer who satisfies the statute without citing it will not show up here. If you know of one we have missed — including your own — send it to hello@trainedon.org and it will be recorded with its source.
What the law asks for
AB 2013 was signed in September 2024 and codified at California Civil Code § 3111. It applies to a developer of a generative AI system or service made publicly available to Californians, for systems released or substantially modified on or after 1 January 2022.
The posted summary must cover, at a high level:
- the sources and owners of the datasets
- how the datasets further the system's intended purpose
- the number of data points, in general categories
- whether the datasets include copyrighted, trademarked or patented material
- whether they include personal information or aggregate consumer information
- whether the developer cleaned, processed or modified the data, and how
- the time period during which the data was collected
- whether synthetic data was used or generated
How it differs from the EU obligation
| California AB 2013 | EU AI Act Article 53 | |
|---|---|---|
| Applies to | Generative AI systems offered to Californians | General-purpose AI models on the EU market |
| In force | 1 January 2026 | 2 August 2025, or 2027 for older models |
| Format | None prescribed | Mandatory Commission template |
| Asks about personal data | Yes, explicitly | Not directly |
| Asks about copyright | Yes, explicitly | Via a separate copyright-policy duty |
A developer can satisfy one and not the other, which is why this register records them as separate fields rather than as one "transparency" verdict. The register shows both per model, and the Article 53 page covers the European side.
Common questions
- What does AB 2013 require?
- Developers of generative AI systems made publicly available to Californians must post, on their own website, a high-level summary of the datasets used to train the system. It covers systems released or substantially modified since 1 January 2022, and it took effect on 1 January 2026.
- What has to be in the summary?
- The sources and owners of the datasets, how they further the system's intended purpose, the number of data points, the types of data, whether the datasets include copyrighted or licensed material, whether they include personal information, whether the developer cleaned or processed the data, the time period of collection, and whether synthetic data was used.
- Is there a template, like the EU one?
- No. The statute prescribes no format and no filing location — the disclosure simply has to be publicly posted. That flexibility is why compliance is hard to survey: there is no register to read and no standard document to look for, which is precisely why this page exists.
- How is it different from the EU obligation?
- The EU requires a summary on a fixed Commission template for general-purpose AI models. California requires a summary in any format for generative AI systems made available to Californians. A provider can satisfy one and not the other, and the two ask for different things — California asks explicitly about personal information and copyrighted material.
- Is it being challenged?
- Yes. xAI has sued the California Attorney General over the statute, arguing among other things that it compels disclosure of trade secrets. The law remains in effect while that is litigated. Notably, xAI has filed training-data summaries under the EU regime — the two positions are not contradictory, but the contrast is instructive about which disclosure obligations providers accept.
Not legal advice. This page describes what the statute says and records what developers have published; whether it applies to you, or has been satisfied, is a question for a lawyer.