Training Data Transparency (AB 2013)

• A federal appeals court is considering a challenge to the California Training Data Transparency Act
• The California law ensures that AI developers disclose the data used to train their systems so the public, regulators, and affected communities can assess whether those systems are trustworthy, lawful, and fair.
• Reporters and researchers also rely on transparency reports to assess AI risks and danger to public safety
•  X.AI claims that the personal data, commercial information, and copyrighted materials it has obtained from others is now X.AI's proprietary information subject to trade secret protection
• X.AI also claims that it should not be required to provide the public basic information about the information it obtained or how it will be used
• Several leading AI companies, including Anthropic and OpenAI, are already working to comply with the law
• Many organizations are planning to file amicus briefs in support of the California Transparency law.
• This page provides background on the case. The amicus briefs are due July 22, 2026. An opinion from the appeals court is expected later this year.

1. What is this case about?

xAI v. Bonta is a lawsuit challenging California’s AB 2013, the Generative Artificial Intelligence Training Data Transparency Act. The law requires developers of public-facing generative AI systems to publish documentation about the data used to train their systems, including information about sources, data types, intellectual property status, personal information, processing, and synthetic data. 

xAI argues that the law violates the US Constitution because it allegedly forces companies to disclose trade secrets, compels speech, and is unconstitutionally vague. California’s position is a permissible transparency measure designed to help the public, regulators, researchers, and affected communities understand how powerful AI systems are developed. 

2. Why is the case significant?

This is one of the first major federal cases testing whether a state can require AI companies to disclose basic information about training data. The outcome could affect not only California’s law, but also future AI accountability laws across the United States.

For the public, the case matters because the choice of training data affects many downstream risks: bias, privacy, copyright, safety, security, explainability, and trust. 

Without basic information about the data used to build AI systems, it is difficult for researchers, policymakers, and courts to evaluate whether these systems are safe, fair, lawful, or reliable.

For AI governance, the case is important because transparency is often the first step toward accountability. A society cannot meaningfully oversee powerful AI systems if the companies that build them can keep all relevant information hidden.

3. What is the current status of the case?

The case was filed by xAI in the U.S. District Court for the Central District of California on December 29, 2025. xAI asked the court for a preliminary injunction, which would have temporarily blocked enforcement of AB 2013 while the case proceeded. The district court denied that request, finding that xAI had not shown it was likely to succeed on the merits. 

xAI then appealed to the U.S. Court of Appeals for the Ninth Circuit. The Ninth Circuit appeal is docketed as No. 26-1591. xAI’s opening brief was filed on May 15, 2026. California will file its response on July 15, 2026. Amicus briefs are due July 22, 2026. 

4. What is a preliminary injunction?

A preliminary injunction is an early court order that temporarily stops a law from being enforced before the court reaches a final decision. Courts usually grant such relief only when the party challenging the law shows a strong likelihood of success and a risk of serious harm.

Here, xAI asked the district court to block AB 2013. The district court declined to do so. The court was not persuaded, at this early stage, that the law should be stopped before the case is fully resolved.

5. What is an amicus brief?

An amicus brief is a “friend of the court” brief. It is filed by a person or organization that is not a party to the case but has expertise, experience, or a broader public-interest perspective that may help the court.

Amicus briefs are especially useful in cases with broad policy consequences. The parties focus on winning the case for themselves. Amici can explain the wider stakes: how a decision could affect public safety, research, consumers, civil rights, innovation, or democratic governance.

6. Why is CAIDP interested in participating?

CAIDP’s interest is straightforward: California’s law is an important effort to promote AI safety, accountability, and public oversight.

CAIDP will explain that AB 2013 is a basic transparency measure. It does not require companies to publish model weights, source code, or every detail of a system’s architecture. It asks for information that helps the public understand the data foundations of generative AI systems made available in California.

CAIDP’s likely focus would be that meaningful AI accountability requires at least three things:

• Transparency about how AI systems are developed.
• Independent evaluation by researchers, regulators, and the public.
• Governance mechanisms that allow safety, rights, and democratic values to be protected before harm occurs.

The CAIDP brief would emphasize that training data transparency is a practical and necessary step toward those goals. Specifically for CBRN (Chemical, Biological, Radiological, and Nuclear) misuse risk, removal of certain types of training data is probably the best safeguard. Without data transparency we cannot tell if this has been done.

CAIDP would build its argument on academic articles, scientific reports, and other objective, evidence-based analysis, including the recent report of the UN Independent International Scientific Panel on AI which found that "Many AI systems that are being used to make decisions

impacting individuals and communities lack sufficient transparency and explainability." The UN expert panel concluded, "There is a critical need for rigorous auditability and transparent data lineage that connects every generated claim back to reliable evidence" and "structural transparency of [AI] systems facilitates public trustworthiness, as it permits external oversight and auditability."

7. Is this case only about California?

No. California is the defendant because AB 2013 is a California law, but the implications are national and potentially global.

Many AI systems are deployed across state and national borders. If California’s law is upheld, it may encourage other jurisdictions to adopt transparency rules. If it is blocked on broad constitutional grounds, it could discourage or limit future efforts to regulate AI through disclosure, documentation, and accountability requirements.

8. Who else is participating as amici?

Several organizations are planning to file amicus briefs in support of the California law, including:

Center for AI and Digital Policy

Center for Investigative Reporting

Knight First Amendment Institute at Columbia University

Legal Advocates for Safe Science and Technology (LASST)

Science Health and Information Clinic at NYU Law (IP scholars)

9. What should I do if I would like to get involved?

Contact Marc Rotenberg at CAIDP.

We are looking for groups interested in filing amicus briefs, as well as First Amendment scholars, IP scholars, and organizations that could be impacted if the California law is defeated.

CONCLUSION

xAI v. Bonta is a major test of whether governments can require basic transparency from AI companies. California’s law asks developers of generative AI systems to tell the public more about the data used to build those systems. xAI says that requirement is unconstitutional. CAIDP’s role would be to help the court understand why transparency is essential to AI safety, accountability, and democratic oversight.