11/20/2026 | Press release | Distributed by Public on 11/20/2026 10:47
A US federal appeals court is weighing Elon Musk's xAI challenge to a California law requiring artificial intelligence companies to disclose summaries of the datasets used to train their models, raising questions about how far governments can go in demanding transparency from technology developers.
A three-judge panel of the 9th US Circuit Court of Appeals in San Francisco appeared skeptical on Wednesday of xAI's argument that the law violates constitutional free-speech protections. However, the judges also questioned whether California had placed clear limits on the amount of information it could compel AI companies to reveal.
The case puts two competing interests at the center of the debate over AI regulation: the public's ability to understand the information used to develop powerful systems and companies' right to protect commercially sensitive methods and trade secrets.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register for Nigeria Capital Market Masterclass.
California's law, signed by Democratic Governor Gavin Newsom in September 2024, requires generative AI companies to publish summaries of datasets used in training their systems. It took effect on January 1 as part of the state's broader effort to establish transparency requirements for the rapidly expanding AI industry.
xAI sued California last year, arguing that the disclosure rules violate its rights under the First Amendment of the US Constitution and could force it to reveal trade secrets about how its models are developed.
The company is seeking to block enforcement of the law while its challenge proceeds. US District Judge Jesus Bernal rejected its request for a preliminary injunction in March, finding that xAI had not demonstrated at that early stage that it was likely to prove the law unconstitutional.
The appeals court's questions on Wednesday suggested some agreement with the lower court's decision, although the judges also appeared concerned about the scope of the state's disclosure powers.
During the hearing, US Circuit Judge Johnnie Rawlinson challenged xAI's argument that the law infringed its free-speech rights, questioning whether requiring companies to disclose information about their training data was an unreasonable government demand.
"In order to say it violates the First Amendment, you have to show that it's not a reasonable request from the government," Rawlinson told xAI attorney James Xi of Clement & Murphy.
"Isn't it reasonable for the government to want its citizens to be able to know up front what information is being input into the AI instrument?" she asked.
The question points to the central justification for training-data transparency rules. AI systems are developed using vast collections of information, and the composition of those datasets can influence how models generate responses, handle sensitive subjects, and reproduce information.
Requiring companies to publish summaries could give users, researchers and regulators a clearer picture of the material behind AI systems. It could also help identify potential problems involving the quality, origins or composition of training data.
For AI developers, however, disclosure requirements raise concerns about how much information must be made public. Details about data selection, sourcing and training processes can form part of a company's competitive advantage. Companies argue that excessive disclosure could expose valuable commercial information to competitors.
The judges appeared to recognize that concern. US Circuit Judge Danielle Forrest asked California attorney Joe Meeker where the limit would be if the state could require companies to disclose information about their training datasets.
"What is the limit?" Forrest asked.
That question could become important to the court's assessment of the law. A requirement to publish broad summaries may be different from an obligation to reveal detailed information about proprietary datasets, internal development processes, or other commercially sensitive material.
The hearing did not produce a ruling, and the court has yet to determine whether the law can remain in force as xAI's constitutional challenge continues.
The dispute carries implications beyond xAI because California's law applies to generative AI companies more broadly. A decision limiting the state's authority could constrain how regulators require AI developers to explain the data behind their systems. A ruling in California's favor could strengthen the legal basis for similar transparency measures.
The case also highlights the difficulty of regulating technology whose development depends on large-scale data collection and processing. Governments face pressure to improve accountability, while companies argue that requirements must not expose proprietary information or impose obligations that are too broad to meet.
The First Amendment argument is another complex variable. xAI is challenging a disclosure requirement on constitutional grounds, rather than simply arguing that compliance would be expensive or technically difficult. Therefore, the appeals court must consider whether the state's transparency objective justifies the speech and disclosure obligations imposed on companies.
For California, the state has the responsibility to establish a workable boundary between information the public has a legitimate interest in accessing and material that companies can reasonably protect as trade secrets.
The case, x.AI LLC v. Bonta, is before the 9th Circuit under docket number 26-1591. Legal experts expect its outcome to help define the limits of state authority over AI development at a time when governments are seeking greater oversight of systems that influence public and commercial life.