California’s generative AI training data disclosure law took effect on January 1, 2026, and it is already before a federal appellate court. On July 16, 2026, the U.S. Ninth Circuit Court of Appeals heard oral argument in X.AI LLC v. Bonta, No. 26-1591, the first appellate test of a state AI transparency mandate anywhere in the country. The outcome will shape not only disclosure obligations for generative AI developers in California, but the constitutional footing of compelled commercial disclosure generally.
What California AB 2013 Requires
The Generative Artificial Intelligence Training Data Transparency Act (“TDTA”), enacted as Assembly Bill 2013 (Irwin) and codified at California Civil Code sections 3110 and 3111, requires a developer of a generative AI system or service made publicly available to Californians to post training data documentation on its website. The obligation reaches systems released on or after January 1, 2022, and applies again upon any substantial modification.
Section 3111 enumerates the categories that the high-level summary must address, including the sources or owners of the datasets, how the datasets serve the system’s intended purpose, the number and types of data points, whether the datasets include personal information or material protected by copyright, trademark, or patent, whether data was purchased or licensed, the time period of collection, and whether synthetic data was used.
Just as important is what AB 2013 does not require. The statute does not compel disclosure of model weights, architecture, source code, training recipes, exact dataset lists, URLs, or the training corpus itself. Certain systems are excluded, including those whose sole purpose is security and integrity, those whose sole purpose is the operation of aircraft in the national airspace, and those provided exclusively for national security or defense purposes.
The Constitutional Challenge
X.AI, the developer of the Grok chatbot, filed suit in the U.S. District Court for the Central District of California against California Attorney General Rob Bonta on December 29, 2025, rather than comply with AB 2013. Other major developers, including OpenAI and Anthropic, posted disclosures on or before the effective date.
The Trade Secret and Takings Theory
X.AI argued in the district court that the compelled disclosures would reveal its trade secrets and therefore effect an uncompensated taking under the Fifth Amendment. On March 4, 2026, Judge Jesus G. Bernal denied X.AI’s motion for a preliminary injunction, finding that its trade secret showing was too generalized. That competitors had complied without demonstrated competitive harm cut against X.AI’s argument. The ruling is a useful reminder that a trade secret objection to a disclosure must be particularized, item by item, and supported by evidence. X.AI has appealed the ruling.
The Compelled Speech Theory
Perhaps X.AI’s stronger argument is that AB 2013 unconstitutionally compels speech. If a training data summary is an ordinary commercial disclosure, it is reviewed under the intermediate speech framework associated with Central Hudson Gas & Electric Corp. v. Public Service Commission, 447 U.S. 557 (1980) and Zauderer v. Office of Disciplinary Counsel, 471 U.S. 626 (1985), and AB 2013 would likely survive scrutiny. If it is compelled non-commercial speech, however, then strict scrutiny applies and the statute very likely falls.
The Central District of California’s decision to deny X.AI’s motion relied on X Corp. v. Bonta, 116 F.4th 888 (9th Cir. 2024), in which the Ninth Circuit struck down portions of a California law (AB 587) requiring social media platforms to disclose content moderation policies to the California Attorney General. The Attorney General’s response at oral argument in X.AI LLC v. Bonta was that a training data summary is closer to a nutrition label than to an editorial position on a contested question.
Why the Stakes Extend Beyond AI
The Zauderer framework underpins mandatory disclosure across financial regulation, environmental reporting, and health and safety law. A holding on appeal in X.AI LLC v. Bonta that factual, categorical training data summaries are not commercial speech would invite challenges well outside the AI sector. An ongoing parallel Ninth Circuit appeal involving California’s climate disclosure statutes raises many of the same questions.
Compliance Steps While the Appeal Is Pending
AB 2013 remains in effect and enforceable during the appeal. Developers therefore face live disclosure obligations regardless of the eventual outcome. Note also that TDTA contains no standalone private enforcement provision; enforcement proceeds through the Unfair Competition Law at the Attorney General’s discretion.
- Determine whether your organization is a developer under TDTA, which can include companies that build internal systems or substantially modify third-party models.
- Inventory and document training datasets against the twelve enumerated categories, and retain the underlying provenance records.
- If asserting trade secret sensitivity as to any category, build a particularized evidentiary record now rather than relying on general assertions.
- Establish a trigger process so that each substantial modification in a dataset or model prompts an updated posting.
Key Takeaway
AI training data transparency is no longer a hypothetical compliance topic in California. It is now an operative legal obligation with an unresolved constitutional overlay, and the Ninth Circuit’s decision in this case may carry consequences far beyond the companies building foundation models.
Speak With a California Business and Technology Attorney
The Law Office of Shanen R. Prout advises technology companies, software and application developers, and content businesses on California regulatory compliance, intellectual property, and commercial litigation. If your company builds or substantially modifies generative AI systems, we can help you assess your obligations under AB 2013.
Shanen R. Prout, Law Office of Shanen R. Prout: 21 Miller Alley, Ste. 210, Pasadena, CA 91103
Telephone: (626) 529-3022
Email: shanen@srplawyer.com
Website: www.srplawyer.com
This article is provided for general informational purposes only and does not constitute legal advice or create an attorney-client relationship. Statutory and case authority should be independently verified, and the status of the pending appeal confirmed, before this article is relied upon.