{"id":108044,"date":"2026-07-16T11:28:16","date_gmt":"2026-07-16T11:28:16","guid":{"rendered":"https:\/\/www.europesays.com\/ai\/108044\/"},"modified":"2026-07-16T11:28:16","modified_gmt":"2026-07-16T11:28:16","slug":"ai-training-data-disclosure-law-faces-first-federal-appeals-court-test","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/ai\/108044\/","title":{"rendered":"AI Training-Data Disclosure Law Faces First Federal Appeals Court Test"},"content":{"rendered":"

California Attorney General Rob Bonta stood before a Ninth Circuit panel on Wednesday and told the judges something that sounds simple but carries enormous weight: requiring AI companies to publish a summary of their training data is no different from requiring a food manufacturer to print a nutrition label. If the Ninth Circuit agrees, California’s AI transparency law survives. If the court disagrees \u2014 if it concludes that training-data summaries constitute something other than ordinary commercial disclosure \u2014 every state AI transparency law in the country faces a constitutional challenge, and the legal framework that permits mandatory business disclosures across finance, environment, and public health could begin to erode with it, Lawfare reported<\/a>.<\/p>\n

The hearing in xAI LLC v. Bonta (No. 26-1591) is the first time a federal appellate court has weighed in on the constitutionality of state-level AI transparency requirements. Every major AI developer operating in California \u2014 and every person who uses generative AI products \u2014 has a stake in the outcome.<\/p>\n

What California’s AB 2013 Actually Requires<\/p>\n

At the center of the dispute is the Generative AI Training Data Transparency Act, enacted as Assembly Bill 2013 and in effect since January 1, 2026. The law requires developers of generative AI systems available to Californians to post a high-level summary of the datasets used to train their systems, Davis+Gilbert explained<\/a>. That summary must address the sources and owners of datasets, whether datasets incorporate personal data or content protected by copyright, trademark, or patent, the approximate size of the datasets, whether the data was purchased or licensed, the time periods over which data was collected, and any modifications made to the datasets, Lawfare noted<\/a>.<\/p>\n

What AB 2013 does not require is equally important: it demands no disclosure of model weights, system architecture, source code, training recipes, exact dataset lists, URLs, data hashes, or the training corpus itself, Jones Walker reported<\/a>. The law also covers AI models released since 2022, even those no longer in active use.<\/p>\n

Several major AI companies, including OpenAI and Anthropic, published the required documentation on their websites by January 1 at varying levels of detail, the Institute for Law & AI noted<\/a>. xAI \u2014 the company behind the Grok chatbot \u2014 is the only major U.S. foundation model developer that filed suit rather than comply, Peter Kahl observed<\/a>.<\/p>\n

Bonta’s Argument: Nutrition Labels for AI<\/p>\n

During Wednesday’s oral argument, Bonta framed AB 2013 as a routine commercial disclosure, telling the appellate panel<\/a> that the law requires AI developers to disclose important factual information that will help consumers decide whether to use a generative AI model relative to other options on the market. He called that “classic commercial speech.”<\/p>\n

The framing is legally precise. Under the Supreme Court’s Central Hudson test and the Zauderer doctrine, compelled commercial disclosures \u2014 factual, non-controversial statements about a product or service \u2014 survive First Amendment scrutiny under intermediate review rather than the more demanding strict scrutiny that applies to compelled ideological expression, AI Challenge Watch explained<\/a>. Requiring a company to state whether its training data included copyrighted material, Bonta argued, is no more constitutionally fraught than requiring a drug company to list ingredients.<\/p>\n

That argument succeeded at the district court level. United States District Judge Jesus G. Bernal denied xAI’s motion for a preliminary injunction<\/a> on March 4, 2026, finding that AB 2013 likely constitutes commercial speech subject to intermediate scrutiny \u2014 and that xAI had failed to demonstrate a sufficient likelihood of success on the merits to justify blocking enforcement.<\/p>\n

xAI’s Three Constitutional Claims<\/p>\n

xAI filed its lawsuit against Bonta on December 29, 2025, raising three constitutional objections. First, that AB 2013 violates the Takings Clause of the Fifth Amendment by compelling disclosure of trade secrets without compensation. Second, that the law forces the company to disseminate specific information against its will \u2014 compelled speech that the First Amendment treats identically to speech restrictions. Third, that the statute’s ambiguous language makes it unconstitutionally vague under the 14th Amendment’s Due Process Clause, as PPC Land detailed<\/a>.<\/p>\n

The trade secret argument is xAI’s most intuitive claim but currently its weakest on the record. Judge Bernal found that xAI offered only generalized assertions that every required disclosure field exposes proprietary strategy and failed to demonstrate with particularity that its datasets are unique trade secrets. The fact that OpenAI and Anthropic have posted similar disclosures without apparent competitive harm substantially undercuts xAI’s position \u2014 if similarly situated competitors can comply without revealing protectable secrets, the burden falls on xAI to explain why its situation differs, Jones Walker noted<\/a>.<\/p>\n

The First Amendment argument is where xAI comes closest to a viable theory, and where the Ninth Circuit’s own recent precedents cut in competing directions. In X Corp. v. Bonta, 116 F.4th 888 (9th Cir. 2024)<\/a>, the Ninth Circuit struck down a California law requiring social media platforms to publicly disclose their content moderation policies, holding that those disclosures compelled non-commercial speech subject to strict scrutiny \u2014 under which the law failed. That ruling is the doctrinal foil xAI invokes. The Ninth Circuit’s Stolfi decision (153 F.4th 795, 9th Cir. 2025) further held that direct public disclosure requirements generally receive strict scrutiny unless they qualify as commercial speech, as Judge Bernal’s order analyzed<\/a>.<\/p>\n

The core question the Ninth Circuit must now answer: is a high-level summary of training-data categories more like a nutritional label (commercial speech, intermediate scrutiny, likely survives) or more like a company’s editorial stance on contested social questions (compelled non-commercial speech, strict scrutiny, likely fails)?<\/p>\n

What Actually Hangs in the Balance \u2014 Beyond AI<\/p>\n

The stakes in xAI LLC v. Bonta extend considerably beyond the survival of one California statute.<\/p>\n

xAI’s constitutional theories, taken to their logical conclusion, would make even minimal AI training-data transparency constitutionally suspect nationwide, Kahl argued<\/a>. The Zauderer commercial disclosure doctrine, which has for decades provided the constitutional foundation for requiring businesses to disclose truthful, factual information in the public interest, was not built around AI. It was built around the same logic Bonta invoked Wednesday: that consumers making commercial decisions are entitled to factual information, and that companies do not have a First Amendment right to withhold it simply because the information relates to their business model.<\/p>\n

If the Ninth Circuit rules that training-data summaries are not commercial speech, it will invite challenges to other disclosure mandates across regulated industries. Legal observers have noted that xAI’s argument, if accepted in broad form, calls into question disclosure mandates across financial, environmental, and health-and-safety regulation \u2014 not just AI transparency requirements in other states and at the federal level, Jones Walker cautioned<\/a>. The outcome of this dispute will reverberate far beyond California, shaping how democracies attempt to regulate frontier AI systems without dismantling the legal foundations of innovation, one analyst concluded<\/a>.<\/p>\n

California is already the most aggressive privacy enforcement jurisdiction in the United States. The California Privacy Protection Agency and the Attorney General collectively issued enforcement actions totaling more than $4.22 million in penalties during the first quarter of 2026 alone, Paul Weiss reported<\/a>. In May, Bonta announced a $12.75 million settlement with General Motors<\/a> to resolve allegations that GM sold Californians’ location and driving behavior data without adequate notice or consent \u2014 the largest CCPA penalty to date.<\/p>\n

Parallel Battles in Other Courts and Legislatures<\/p>\n

The xAI v. Bonta hearing does not stand alone in the Ninth Circuit’s disclosure-law docket. In January 2026, a three-judge panel heard oral argument<\/a> in a separate case challenging California’s Climate Corporate Data Accountability Act and Climate-Related Financial Risk Act \u2014 both of which require companies to disclose emissions data and climate risk assessments. The panel in that case probed many of the same First Amendment frameworks around compelled speech and commercial speech status.<\/p>\n

xAI has also taken its constitutional theory to Colorado, filing a complaint in April 2026<\/a> against Colorado Attorney General Philip Weiser seeking to block enforcement of the Colorado AI Act. That case is pending in federal court. Colorado subsequently replaced its original AI law with a narrower disclosure statute that takes effect January 1, 2027 \u2014 though the federal litigation over the original law remains active.<\/p>\n

At the federal level, a bipartisan legislative discussion draft released in June 2026 \u2014 the Great American Artificial Intelligence Act \u2014 would preempt state AI development laws including AB 2013 for three years, while imposing federal third-party audit requirements on large AI developers. That draft has not been enacted.<\/p>\n

Meanwhile, the European Union’s AI Act requires foundation model providers to publish training-data summaries to downstream users as one of its general-purpose AI obligations, with enforcement authority and potential fines beginning August 2, 2026.<\/p>\n

California’s Deepening AI Accountability Apparatus<\/p>\n

The appellate hearing arrives as California operates a multi-pronged AI and privacy enforcement apparatus that is expanding in both scope and financial consequence.<\/p>\n

Starting August 1, 2026, data brokers must process consumer deletion requests through the state’s centralized Delete Request and Opt-Out Platform \u2014 with fines of $200 per deletion request per day<\/a> for failure to comply. More than 300,000 California residents have already registered on the platform and submitted deletion requests since it launched in January 2026.<\/p>\n

Attorney General Bonta has simultaneously built an AI accountability program to strengthen oversight while federal AI governance remains in flux, leaving California and other states as the primary rule-writers for the foreseeable future, Jones Walker noted<\/a>. The same office defending the training-data transparency law is also investigating xAI over its generation of alleged non-consensual imagery \u2014 a separate proceeding illustrating how quickly AI governance questions are converging on the same companies from multiple directions.<\/p>\n

AB 2013 itself lacks a standalone enforcement provision; enforcement runs through California’s Unfair Competition Law at the AG’s discretion. That enforcement gap means the statute’s practical force depends significantly on how aggressively Bonta’s office deploys it \u2014 which, given Wednesday’s courtroom appearance, appears to be aggressively.<\/p>\n

Ninth Circuit Ruling: What to Expect<\/p>\n

The Ninth Circuit panel typically deliberates for several months following oral argument. No ruling is expected imminently. AB 2013 remains in effect and enforceable while the appeal proceeds, meaning AI developers operating in California face live disclosure obligations regardless of the eventual outcome.<\/p>\n

If the Ninth Circuit finds that high-level training-data summaries constitute commercial speech subject to intermediate scrutiny, AB 2013 will almost certainly survive \u2014 California would simply need to show that the law directly advances a substantial government interest and is no more extensive than necessary, a test consumer-transparency laws routinely clear. If the court instead rules that the summaries amount to compelled non-commercial speech \u2014 applying the framework from X Corp. v. Bonta \u2014 AB 2013 would face strict scrutiny, a standard disclosure laws rarely survive.<\/p>\n

Judge Bernal noted in his March ruling that even if the Ninth Circuit affirms the denial of xAI’s preliminary injunction, xAI retains a distinct possibility of prevailing on the merits as the case continues to develop, Lawfare reported<\/a>. The constitutional question remains genuinely open. For any organization that builds, deploys, or relies on generative AI products in California, the Ninth Circuit’s answer will determine whether knowing what went into an AI system is something you are legally entitled to \u2014 or something the company has a constitutional right to withhold.<\/p>\n

California’s Delete Act Adds Another August Compliance Deadline<\/p>\n

The xAI v. Bonta appellate hearing lands less than three weeks before August 1, 2026 \u2014 when the most consequential phase of California’s Delete Act takes effect. On that date, every registered data broker in the state must begin checking the state’s centralized deletion platform every 45 days and process consumer deletion requests, or face fines of $200 per deletion request per day<\/a>. The California Privacy Protection Agency had already fined multiple unregistered data brokers in 2026, and the agency’s data broker enforcement strike force<\/a> launched in late 2025 continues to expand its footprint.<\/p>\n

For AI developers whose training data includes consumer information \u2014 a category AB 2013’s disclosures would document \u2014 the Delete Act’s new compliance requirements create an intersecting legal obligation: not just to disclose what data was used, but to respond when consumers request its deletion from their systems.<\/p>\n

Frequently Asked QuestionsWhat does California’s AB 2013 require AI companies to disclose?<\/p>\n

AB 2013 requires developers of generative AI systems available to Californians to post a publicly accessible summary covering the sources and owners of their training datasets, whether those datasets include copyrighted, trademarked, or personal information, the approximate size of the data, whether it was purchased or licensed, and the time span of data collection. The law does not require disclosure of model weights, architecture, source code, or the actual training corpus \u2014 only a high-level categorical summary, as Davis+Gilbert explained<\/a>. The law retroactively covers models released since 2022.<\/p>\n

Why is xAI the only major AI company fighting AB 2013 in court?<\/p>\n

OpenAI and Anthropic both posted AB 2013-compliant disclosures on or before January 1, 2026, apparently without revealing information they considered competitively sensitive, Peter Kahl noted<\/a>. xAI has argued that its specific dataset curation methods constitute trade secrets that compliance would expose, but the district court found those allegations too general to support the claim \u2014 noting that competitors complied without evidence of competitive harm. xAI’s broader constitutional challenge also targets the legal doctrine that permits any factual commercial disclosure requirement, not just this specific law.<\/p>\n

What happens to AI transparency laws nationwide if xAI wins at the Ninth Circuit?<\/p>\n

If the Ninth Circuit rules that training-data summaries constitute compelled non-commercial speech subject to strict scrutiny \u2014 rather than commercial disclosure subject to intermediate scrutiny \u2014 the ruling would extend well beyond AB 2013. Legal analysts note that accepting xAI’s argument in broad form would call into question the constitutional framework supporting disclosure mandates across financial regulation, environmental reporting, and health-and-safety law, Jones Walker cautioned<\/a>. Other states with similar AI training-data transparency bills would face immediate constitutional pressure, and federal AI transparency proposals would confront the same doctrinal challenge.<\/p>\n

As someone who uses AI tools, do I have a legal right to know what data trained them?<\/p>\n

Under AB 2013, California residents currently have an indirect right: developers must post the required summary publicly, and anyone can read it. That right exists because the law is in effect while the appeal proceeds. Whether it remains in place after the Ninth Circuit rules depends on the court’s First Amendment analysis. No federal equivalent exists yet, though the Great American AI Act discussion draft would impose federal audit requirements on large AI developers. The EU AI Act’s general-purpose AI provisions require training-data summaries to be provided to downstream business users as of August 2, 2026, though not to individual consumers directly.<\/p>\n","protected":false},"excerpt":{"rendered":"California Attorney General Rob Bonta stood before a Ninth Circuit panel on Wednesday and told the judges something…\n","protected":false},"author":2,"featured_media":108045,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[10],"tags":[4788,55864,55862,26411,11760,33353,2899,55863],"class_list":["post-108044","post","type-post","status-publish","format-standard","has-post-thumbnail","category-xai","tag-ai-regulation","tag-ai-training-data-transparency","tag-california-ab-2013","tag-california-privacy-law","tag-first-amendment","tag-ninth-circuit","tag-xai","tag-xai-v-bonta"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/108044","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=108044"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/108044\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/108045"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=108044"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=108044"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=108044"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}