The Justice Department told a federal judge that Minnesota's nudification ban sweeps too broadly, without joining the case.

On August 18 the United States told a federal court that a Minnesota law goes too far — in a case the United States has not joined. The law took effect on August 1 and bans operating a nudification service: one that alters a photograph of a real, identifiable person to show a body part the original did not.

The vehicle is a statement of interest under 28 U.S.C. § 517, which lets the Attorney General send a Justice Department lawyer to address a court in a pending case without the government becoming a party. It is a lighter instrument than intervention, and the filing says so: the United States “reserves all rights,” and observes in a footnote that the Attorney General could intervene to vindicate federal supremacy. Signed by Associate Attorney General Stanley E. Woodward, Jr., it lands in X.AI LLC v. Ellison, where Elon Musk’s xAI is arguing that HF 1606 violates the First Amendment.

Read closely, the objection is narrower than the framing suggests, and more technical. The Justice Department’s objection is that Minnesota wrote strict liability where the comparable federal statutes require that an act be done knowingly, gave platforms no notice-and-removal window of the sort the TAKE IT DOWN Act provides before liability attaches, and pulled its definition of “intimate part” out of a criminal statute written about physical contact. That last borrowing is what produces the filing’s own illustration: HF 1606 “would apply to an artificially generated image of a shirtless man in a swimming pool,” where “federal law does not.” xAI’s reply brief makes the commercial version of the point, that the same edit revealing an inner thigh or a bare chest exposes the provider to a $500,000 penalty — a figure the statute sets per unlawful access, download or use.

What makes this more than a drafting quarrel is the authority the filing cites for why breadth matters. It quotes Executive Order 14365 for the proposition that it is “the policy of the United States to sustain and enhance the United States’ global AI dominance through a minimally burdensome national policy framework for AI” — one “national standard,” rather than “50 discordant state ones” — and argues HF 1606 “stretches far more broadly with its strict liability regime to risk paralyzing the AI industry and hampering United States leadership in AI.” That is the executive order’s preemption strategy being run through a district court instead of through Congress, and the test case is a child-protection statute the legislature passed nearly unanimously. The same filing cites a second order, the one that put a 30-day federal review in front of frontier model releases, for the federal interest it says the state law disturbs. Minnesota’s opposition brief leans on exactly that, citing Moody v. NetChoice for the cost a plaintiff accepts in bringing a facial challenge. xAI is separately a defendant in a California class action alleging Grok produced child sexual abuse material from real childhood photographs.

Update, October 8, 2026: Judge Frank denied the preliminary injunction on September 4, concluding that xAI had unreasonably delayed and had not shown irreparable harm, and refused an injunction pending appeal on September 9. The Eighth Circuit’s October 2 order is a single sentence: “Appellant’s motion for injunction pending appeal is granted.” At both parties’ request, the motion to dismiss waits until after the appeal, and its November 5 hearing was struck. The court amended the caption on October 8 to reflect that X.AI LLC is now known as SpaceXAI LLC.

For anyone building image tools with US users, the operative fact is the timetable rather than the constitutional argument: a strict-liability state exposure took effect on August 1, and the federal government’s view of it carries no legal weight until a judge adopts it. The compliance work that survives either outcome is the same work platforms are already doing for Amazon’s synthetic-performer tagging rules — knowing what an edit did to an image, and being able to show it.

Related

• The White House says Moonshot AI trained its model on Anthropic’s, and accessed restricted Nvidia chips in Thailand
• Amazon now wants a hidden metadata tag on every photorealistic AI face in your product images
• The government now gets a look at the most powerful AI models before the public does