Summary
California Assembly Bill 2013, the Generative AI Training Data Transparency Act, took effect January 1, 2026. It is the first U.S. statute to require AI developers to post a high-level summary of the datasets used to train each generative AI system. The statute is enforced by the California Attorney General; the operative provisions sit in the California Business and Professions Code.
What it requires
Developers of generative AI systems made available to Californians on or after January 1, 2026 must post on their public websites a high-level summary of the datasets used to develop each system. The summary must address, at minimum:
- The sources or owners of the datasets.
- A description of how the datasets further the system's intended purpose.
- The number of data points in the datasets.
- Whether the datasets include data protected by copyright, trademark, or patent, or are entirely in the public domain.
- Whether the datasets include personal information as defined by the CCPA, or aggregate consumer information.
- Whether any cleaning, processing, or labeling occurred on the datasets.
- The time period during which the data was collected, including a notice if the collection is ongoing.
Who it covers
The statute applies to developers of generative AI systems — defined to include the entities that develop or substantially modify the system — that are made available to Californians. It exempts certain narrow categories (national-security applications, certain medical-device-cleared systems) and treats the disclosure obligation as a continuing duty: when datasets change materially, the disclosure must be updated.
Litigation
xAI v. Bonta — constitutional challenge
U.S. District Court · PendingxAI Corp has sued California Attorney General Rob Bonta to enjoin enforcement of AB 2013. The complaint reportedly raises two principal theories: that the statute compels speech about training datasets in violation of the First Amendment, and that it regulates extraterritorial AI development in violation of the dormant Commerce Clause. The filing district has been reported as the Eastern District of California; the docket has not been publicly confirmed in sources reviewed.
Source: IAPP — xAI v. Bonta: a constitutional clash for training-data transparency.
Why it matters
- It moves disclosure from a copyright remedy to a statutory duty. Most U.S. AI training-data fights have been waged through copyright litigation. AB 2013 creates a freestanding state-law obligation that does not require a copyright owner as plaintiff.
- It creates discoverable inputs. The published summaries become the natural starting point for plaintiffs in copyright, privacy, and consumer-protection actions.
- It tees up a federal-preemption question. Federal AI legislation, if it arrives, will have to decide whether to preempt state transparency statutes; the xAI challenge is the early stress test.
FAQ
When did AB 2013 take effect?
January 1, 2026.
Does it apply to open-source model developers?
The statute reaches developers who substantially modify a covered model. Open-source releases that are made available to Californians fall within the same disclosure framework; specific edge cases (e.g., noncommercial research releases) are likely to be litigated.
How is it enforced?
Enforcement is by the California Attorney General. The statute does not, on its face, create a private right of action.
What's the relationship to AB 2885 and SB 1047?
AB 2013 is part of a broader package of California AI bills. SB 1047 (the safe-AI bill) was vetoed in 2024; AB 2885 addresses generative AI definitions. AB 2013 is the live transparency statute.