Two lawsuits. Two state attorneys general. One constitutional theory that could reshape AI governance across the United States.
xAI, Elon Musk’s AI company, is simultaneously appealing a federal court loss in California and pressing a new complaint against Colorado’s AI regulators. The company’s core argument is the same in both states: laws that compel AI developers to disclose information about their training data violate constitutional protections for speech, property, and equal protection. If that argument succeeds at the 9th Circuit, the federal appellate court with jurisdiction over California, and whose precedents carry significant weight nationally, the legal viability of AI transparency regimes built on disclosure mandates becomes genuinely uncertain.
Here’s who’s in the arena, what each party is arguing, and what the decision tree looks like.
The Two Fronts: California and Colorado
California’s AB 2013 took effect January 1, 2026. It requires AI developers to publish high-level summaries of the data used to train their systems, not the data itself, but documentation of its sources, categories, and scope. xAI sued California Attorney General Rob Bonta, arguing the law violated the Constitution. The Central District of California denied xAI’s motion for a preliminary injunction on or around March 4-5, 2026, per court filings in Case 2:25-cv-12295-JGB-SSC reviewed by Jones Walker. The denial means AB 2013 enforcement continues while xAI appeals to the 9th Circuit.
Colorado’s AI Act (SB 24-205) is a broader law than AB 2013, covering algorithmic accountability and discrimination risks in addition to transparency. It’s also unsettled: local legislators are still debating amendments, according to HRDive. xAI filed its complaint against Colorado AG Philip Weiser on April 9, 2026, per IAPP’s coverage. That complaint raises the same constitutional grounds as the California case.
The Constitutional Arguments
xAI’s lawsuits assert three constitutional theories. Each one targets a different vulnerability in how disclosure mandates are structured.
First Amendment, Compelled Speech. The most developed argument. xAI contends that requiring AI developers to disclose information about their training practices is compelled speech, the government forcing a private party to say something it would rather not say. The compelled speech doctrine has a well-established body of First Amendment case law; the question is whether it extends to commercial disclosures about AI training methodology. That question has no appellate precedent. No circuit court has ruled on whether AI training data disclosures qualify as speech protected from government compulsion. This is xAI’s argument, not an established legal rule.
Fifth Amendment, Taking. The complaint is reported to allege that mandatory disclosure of training data constitutes a governmental appropriation of trade secrets without just compensation, a “taking” under the Fifth Amendment’s Takings Clause, according to IAPP’s reading of the complaint. The specific allegation text hasn’t been confirmed from primary court documents. If the court accepts this framing, it would mean states could only require such disclosures if they compensated developers for the competitive value of the disclosed information. That’s a high bar.
Fourteenth Amendment, Equal Protection. The equal protection argument is least developed in available reporting. Its inclusion suggests xAI may be arguing the laws treat AI developers differently from similarly situated parties without sufficient justification.
The States’ Position
California and Colorado enacted their laws for distinct but related reasons. AB 2013’s drafters argued that AI developers hold significant information asymmetry over the public, users and regulators can’t assess AI system risks without knowing something about what the system was trained on. Transparency is the mechanism for reducing that asymmetry.
Colorado’s AI Act goes further, tying disclosure to accountability for discriminatory algorithmic outcomes. The state’s position is that disclosure requirements are factual and non-ideological, they compel facts, not opinions, and should therefore receive deferential rather than heightened constitutional scrutiny. That’s the legal line the states need to hold: if courts treat these disclosures as purely factual commercial reporting requirements, the First Amendment argument weakens substantially.
The compelled speech doctrine has generally distinguished between ideological speech (highly protected) and commercial factual disclosures (lower protection). The states will argue training data summaries are closer to nutrition labels than political manifestos. xAI will argue that training data selection is expressive, a creative editorial process, and therefore closer to protected speech.
The Domino Scenario
The 9th Circuit covers California, Oregon, Washington, Arizona, Nevada, Montana, Idaho, Alaska, and Hawaii. A ruling striking down AB 2013 would apply directly across those states. More significantly, 9th Circuit opinions carry persuasive weight in other federal circuits, they’re not binding outside the 9th, but they’re cited, and they shape how attorneys advise their clients.
Legal analysts and state lawmakers in multiple states are watching the 9th Circuit closely. The specific downstream impact on pending AI legislation elsewhere remains uncertain, but the pattern is visible. Several states modeled their AI transparency bills partly on California’s framework. A constitutional ruling against that framework would generate immediate pressure to redesign or withdraw similar bills.
What Compliance Teams Need to Know Right Now
The injunction denial has a practical implication most headlines miss: AB 2013 is in force. The appeal doesn’t automatically stay enforcement. xAI may seek an emergency stay from the 9th Circuit, but no such stay has been reported. Until and unless the 9th Circuit orders otherwise, AI developers subject to AB 2013 must comply.
What that means operationally: if your organization deploys AI systems that meet AB 2013’s applicability thresholds, and those thresholds cover AI systems used in substantive decisions, you need training data documentation in place now, not after the appeal resolves. A 9th Circuit decision is likely 12-24 months out. That’s two compliance cycles where AB 2013 governs.
Colorado is a different calculus. The law is contested and still being amended. Monitor the Colorado legislative calendar alongside the xAI complaint. If the legislature significantly amends SB 24-205 before the court rules, the lawsuit may need to be refiled against the amended version.
TJS Synthesis
xAI is doing something more strategic than fighting two disclosure requirements. It’s building the appellate record that could make AI training data transparency laws legally precarious everywhere. The company has the resources for a prolonged 9th Circuit battle, and the constitutional questions it’s raising are genuine, courts haven’t answered them.
The compliance implication is real and immediate: organizations relying on a future favorable ruling to defer AB 2013 compliance are accepting regulatory risk they don’t need to accept. The law is in force today. Build the documentation. Watch the 9th Circuit docket. And prepare to adapt your compliance posture if the appellate court produces a result that surprises in either direction.