
California’s new AI disclosure law faces a fierce First Amendment and property-rights challenge from Elon Musk’s xAI in the Ninth Circuit.
Story Highlights
- xAI argues California’s AB 2013 compels speech and risks exposing trade secrets.
- The Ninth Circuit heard xAI’s appeal after a trial judge denied an injunction.
- Judges pressed xAI as California framed the law as simple transparency.
- AB 2013 requires high-level training-data summaries for public-facing AI.
What The Case Is About: Speech, Secrets, And State Power
xAI told the court that California’s AB 2013 forces it to publish model-specific training-data details, which it says is compelled speech that fails strict scrutiny under the First Amendment. The company also warned that the law could reveal “fiercely protected” trade secrets without pay, which it argues would be an unconstitutional taking of property. California says the law is only about transparency for users, not about forcing a viewpoint or exposing code or datasets.
California’s statute requires developers to post a high-level summary of the datasets used to train public-facing generative systems, including sources, data types, size, intellectual-property status, licensing, personal information, processing, collection period, and use of synthetic data. Legislative analyses describe these disclosures as summaries, not full datasets or source code. Those materials pitch the law as helping people compare systems and judge reliability before they trust them.
Where The Appeal Stands: A Skeptical Panel And An Active Docket
The United States Court of Appeals for the Ninth Circuit is reviewing xAI’s appeal after a district court refused to halt the law while the case proceeds. News reports said the panel pressed xAI’s arguments and signaled doubt about blocking the law at this stage. During arguments, Judge Johnnie B. Rawlinson asked whether it was reasonable for the state to let people know what goes into an AI system, pressing xAI’s view of commercial-speech limits.
xAI filed its opening brief on May 15, 2026, in case number 26-1591, framing AB 2013 as content- and viewpoint-based compelled speech that burdens its models and messaging. An amicus brief backing California supported the trial court’s denial of an injunction and stressed the public’s interest in basic training-data facts to evaluate claims about safety, bias, and reliability. The law took effect January 1, 2026, so disclosures are required unless the court steps in.
What AB 2013 Demands: High-Level, But Still A Mandate
Legislative documents say AB 2013 covers generative systems offered to Californians, including free tools, and requires a high-level summary of training datasets posted on a developer’s site. The list of fields is long and specific, reaching source or owner, purpose, scale, data types, intellectual-property status, licensing, personal information, processing, time frames, and synthetic-data use. The statute also asks developers to explain how datasets serve the system’s stated purpose, tying disclosures to function and design.
California frames these as factual, noncontroversial statements about a commercial product, which courts often review under a more forgiving standard for disclosures about goods and services. xAI argues that these fields still reveal core choices that define a model’s value, strategy, and supplier ties, which the company views as protected expression and property. The current record does not show a specific trade secret that must be named, which limits how far that claim can go today.
Why Conservatives Should Care: Free Speech, Innovation, And State Overreach
This fight is bigger than one company. If California can force speech about how models are built, other states may follow with longer checklists and shifting rules. Developers could face a patchwork of mandates, higher costs, and pressure to reveal competitive playbooks. That path chills speech, weakens property rights, and hands more leverage to bureaucrats who want to micromanage tools people use every day. The Constitution guards against that kind of compelled speech and uncompensated takings for good reason.
Supporters of AB 2013 say the law is just transparency. But the government’s “just tell us more” rarely stops at a summary. Even high-level fields can map out where a company gets data, how it filters content, and what it values. That is the recipe that drives innovation and jobs. Courts will decide which standard applies, but the stakes are clear: keep speech and property protections strong, or invite a new era of tech policy by press release from Sacramento.
Sources:
law.com, motherjones.com, mlex.com, law360.com, iapp.org, caidp.org, sjud.senate.ca.gov, apcp.assembly.ca.gov, leginfo.legislature.ca.gov, crowell.com














