xAI Beats Minnesota: 8th Circuit Freezes AI Nudification Ban

By Daniel Rao

Photo: Elon Musk at the 2015 Tesla Motors annual meeting. His company xAI persuaded the 8th US Circuit Court of Appeals to put Minnesota's first-in-the-nation AI "nudification" ban on hold. (Steve Jurvetson, via Wikimedia Commons, CC BY 2.0)

The Minnesota legislature passed its AI "nudification" ban by 132 votes to one. Two months after it took effect, a federal appeals court has frozen it — at the request of Elon Musk's company, which says the law is unconstitutional.

That is the kind of sentence that should make both sides uncomfortable, and it does. The 8th US Circuit Court of Appeals, based in St. Louis, granted xAI's request for an injunction on Friday, 2 October, putting the law on hold "for now" while the company pursues its lawsuit. Minnesota's attorney general had urged the court not to. The whole fight is over a statute that was supposed to be the easy part of the AI debate — the bit where everyone agrees fake nudes of real people are bad — and it is now a First Amendment knife fight with $500,000 fines on the table.

I've covered a lot of AI regulation this year. This is the first one where the winning argument might be: the law works so well it can't be allowed to work at all.

The law that scared a billionaire

The statute, known as HF 1606, is the first of its kind in the country. It does not target the people who generate fake nude images. It targets the operators — website owners, app makers, software developers — that let users create realistic images adding intimate body parts to photos of identifiable people. The penalties are genuinely severe: Minnesota's attorney general can seek civil fines of up to $500,000 for each unlawful access, download or use, with the money going to victim-services grants. On top of that, any person depicted in such an image gets a private civil action — up to three times actual damages, plus punitive damages, injunctive relief and attorney's fees.

The detail that matters most, and the one xAI is actually fighting about, is what the law does not require. There is no knowledge requirement. Under the federal Take It Down Act and a 2025 Texas deepfake law, a provider has to know the image was nonconsensual before liability attaches. Minnesota removed that condition entirely. A provider can be on the hook even without awareness that a nude image was generated. It is strict liability, applied to AI image generators, and nothing else in American law looks like it.

The bill's politics were overwhelming. Governor Tim Walz signed it on 7 May after the House passed it 132–1 and the Senate 65–0. The trigger, as reported at the time, was a man who used social-media photos to fabricate sexual images of more than 80 women he knew. When a law passes by those margins, it usually means the hard questions were postponed. Minnesota is discovering that postponement has a half-life.

What the lower court thought of xAI's rush

Before the appeals court intervened, xAI was losing this fight badly. US District Judge Donovan Frank denied a temporary restraining order on 31 July — a day before the law took effect — and then denied a preliminary injunction on 4 September. His reasoning had a lawyer's precision and a judge's impatience. xAI, he found, had not shown it would suffer irreparable harm while challenging the law. And the company had, in his view, waited too long to ask.

"If xAI genuinely feared irreparable harm, the Court does not doubt it would have acted more quickly to file this action and move for injunctive relief," Frank wrote, as reported by Reuters. It is the kind of sentence that lands in legal textbooks under "what not to do in emergency litigation." Frank also found xAI's evidence of added costs and commercial injury too vague to justify blocking the statute.

We ran the calendar on xAI's timeline, and Frank's irritation becomes understandable. Walz signed the bill on 7 May. xAI sued Attorney General Keith Ellison on 27 July — eleven weeks later — and asked for emergency relief two days after that. A company racing to avoid catastrophe does not wait until the eve of enforcement to ask a court for help. xAI's answer, in its appeal, was that "large corporate entities may have greater difficulty in moving quickly because of the number of decision-makers involved and the varied interests a large corporation must balance." Read that twice. It is the legal equivalent of "the dog ate my injunction."

The district court then denied xAI's request for an injunction pending appeal on 9 September. So Friday's order from the 8th Circuit is not a change of heart — it is a different court reaching a different conclusion about the same emergency.

Why Friday's order changes nothing — and everything

Here is what the appeals court's order is not: a ruling on the merits. Nobody at the 8th Circuit has decided whether Minnesota's law violates the First Amendment. The order puts enforcement on hold while xAI's appeal proceeds. The underlying lawsuit — xAI's claim that the measure restricts constitutionally protected free speech — continues in the district court, where Minnesota's August motion to dismiss is on hold pending the injunction appeal.

Here is what it is: the first appellate weighing-in on the first law of its kind, and a signal that the company's constitutional arguments are at least serious enough to pause enforcement. When a federal appeals court enjoins a democratically enacted statute, even temporarily, it is saying the challenger might win. That "might" now hangs over every similar bill in every other statehouse.

Photo: The Minnesota State Capitol in St. Paul. HF 1606 passed the House 132–1 and the Senate 65–0 before Governor Tim Walz signed it in May. (xiquinhosilva, via Wikimedia Commons, CC BY 2.0)

Update, Oct. 3: since publication, Reuters (via CNA) has reported the first official reactions to Friday's freeze. Attorney General Keith Ellison's office said it was "disappointed by the appeals court ruling and will defend the new state law." Ellison's spokesperson John Stiles defended the statute directly: "Minnesota's nudification ban outlaws AI technology products from generating sexual images that harm and harass people in the vilest way possible." Reuters also confirmed that xAI "has begun suing users whom it alleges are evading Grok's technological blockers to create sexual images of people without their consent" — the company's parallel strategy of policing its own users while asking courts to bar the state from policing it. xAI and Minnesota did not otherwise comment.

The First Amendment question nobody can dodge

xAI's core argument is that HF 1606 restricts speech the Constitution protects. The company says its Grok Imagine tool already has "rigorous protections" against creating nudified or sexualised images of real people, that it strictly prohibits users from generating non-consensual nude or sexualised images, and that it has imposed technological barriers to prevent such images from being created. In court filings, xAI even proposed suspending features for Minnesota users rather than risking the fines — an alternative the district court found insufficient.

The company is represented by Robert Dunn of the law firm Eimer Stahl, a former clerk to US Supreme Court Justice Clarence Thomas. That is not a detail to skip past. You do not hire that résumé for a routine commercial dispute. You hire it because you expect to be making First Amendment arguments in front of judges who care about them, possibly all the way up.

Minnesota's counter is blunt and political. The state adopted the measure, it told the appeals court, to combat the spread of what it called a "staggering amount (of) child sexual abuse material that AI products like Grok Imagine generate." Attorney General Keith Ellison, who has defended the law in court, did not sound like a man hedging his bets after the September ruling: "I am extremely proud to be defending this law, and along with it, the dignity of the people of Minnesota," he said in a statement. "These nudification apps have been used to generate child sexual abuse materials and harass people in the vilest ways imaginable," he added. "That repulsive behavior is not welcome in Minnesota."

Ellison's statement carried an unusual note of open anger for an attorney general's office. Most AG press releases read as if a committee wrote them. This one reads as if a person did.

The uncomfortable truth for both sides is that the strongest arguments are aimed at different targets. The strict-liability design really is radical — it makes Minnesota's law broader than the federal statute Congress passed for the same problem, and broader laws are harder to defend against overbreadth challenges. But "rigorous protections" is doing enormous work in xAI's filings, and it is doing it in the shadow of some grim reported numbers. The same appetite for harder edges is showing up in Washington, where senators Hawley and Murphy want prison time for AI companies whose agents hack systems.

Grok's own numbers are xAI's worst enemy

According to figures cited in the district court proceedings and reported by BigGo Finance, Grok produced approximately 3 million sexualised images in the 11 days after a new image-editing feature launched — including about 23,000 that appeared to depict children. Those are the kinds of numbers that end careers in most industries. xAI's defence, as reported, is that it suspended 52,222 accounts this year and filed 73,604 reports with the National Center for Missing and Exploited Children.

Both numbers can be true at once. A system can have "rigorous protections" and still generate millions of sexualised images, because scale is the whole point of the product. That is the paradox the First Amendment argument has to survive: xAI is asking a court to protect its speech rights while its own data, reportedly, shows the speech in question is being generated by the truckload — some of it depicting children. It is the same pattern of agents and tools misbehaving at scale that saw OpenAI's own agents caught casing US government systems, and it keeps landing in courtrooms faster than legislatures can write rules for it.

The company, to its credit, is not pretending the abuse problem is someone else's. xAI has begun suing users it alleges are evading Grok's technological blockers to create sexual images of people without their consent. There is something almost admirable in the symmetry: Musk's company wants the state of Minnesota barred from policing its product, while it polices its users itself. Whether you call that principled or convenient probably depends on how much you trust Elon Musk with a content-moderation apparatus — and the answer to that question is doing a lot of quiet work in this case.

Minnesota is not the only jurisdiction moving against AI harms to minors — Australia just named and shamed the worst kid-safety offenders in gaming.

Minnesota's nudification ban: the questions everyone asks

Is Minnesota's ban in effect right now?
No. On Oct 2, the 8th US Circuit Court of Appeals froze enforcement of HF 1606 while xAI's appeal proceeds. That reverses the lower court's stance: US District Judge Donovan Frank denied xAI a temporary restraining order on 31 July, a preliminary injunction on 4 September, and an injunction pending appeal on 9 September.

What exactly did the 8th Circuit order?
A temporary freeze on enforcement pending the appeal — not a ruling on the merits. Nobody at the appeals court has decided whether the law violates the First Amendment; the underlying lawsuit continues in the district court, where Minnesota's August motion to dismiss is on hold pending the injunction appeal.

What is xAI's First Amendment argument?
That HF 1606 is a content-based restriction on constitutionally protected speech: the company says Grok Imagine has "rigorous protections" against nudified images of real people, strictly prohibits non-consensual sexualised images, and has technological barriers in place — and that the law's strict-liability design (no knowledge requirement, unlike the federal Take It Down Act or Texas's 2025 deepfake law) makes it vulnerable to overbreadth challenges. xAI is represented by Robert Dunn of Eimer Stahl, a former clerk to Supreme Court Justice Clarence Thomas.

What happens next — and what did the AG say?
Three things: the 8th Circuit's full decision on the injunction appeal, Minnesota's frozen motion to dismiss, and xAI's parallel lawsuits against users it says evade Grok's blockers. Attorney General Keith Ellison's office said it was "disappointed by the appeals court ruling and will defend the new state law." His spokesperson John Stiles said the ban "outlaws AI technology products from generating sexual images that harm and harass people in the vilest way possible."

Our take: the law is botched, the lawsuit is cynical, and both can be true

Let me say the unfashionable thing. HF 1606 is badly drafted law. Strict liability for AI image generation — no knowledge requirement, $500,000 per violation, private treble-damage actions — is a legislative sledgehammer, and sledgehammers are how you get constitutional challenges you lose. The bill passed 132–1 precisely because nobody wanted to be the one person voting against a fake-nudes ban, which is the worst possible reason to skip the drafting discipline. Minnesota's legislature wrote a law designed to be applauded, not a law designed to survive Robert Dunn.

And xAI's lawsuit is a cynical exercise dressed as a principle. The First Amendment argument is real — content-based restrictions on image generation genuinely raise constitutional questions, and the strict-liability feature makes those questions harder. But the company waited eleven weeks to sue, proposed a Minnesota-only feature suspension as its grand alternative, and is simultaneously in court arguing it polices this exact content just fine on its own. That is not the posture of a company defending free expression. It is the posture of a company defending its margins, and $500,000-per-violation margins are the kind that focus the mind.

Here is what will decide the case, and probably the next five years of AI regulation: whether the courts believe a state can make AI companies strictly liable for what their users generate. If the answer is yes, every statehouse in America will copy Minnesota by spring. If the answer is no, the regulatory ceiling for AI image tools is whatever the companies promise — and we have all seen how those promises hold up at scale. The UK's AI safety institute has just restarted cyber evaluations after its own test agents went rogue; the people best equipped to contain these systems keep discovering the limits after the fact.

What we're watching next

Three things. First, the 8th Circuit's full decision on the injunction appeal — Friday's order is temporary, and the court's reasoning, when it arrives, will be the first appellate signal on the constitutional merits. Second, Minnesota's motion to dismiss, frozen since 1 September, which will test whether the case even survives to a full hearing. Third, and most interesting: whether xAI's parallel strategy of suing its own abusive users produces anything enforceable, or whether it exists mostly as an exhibit in the First Amendment brief.

The honest headline out of St. Louis this week is not that Musk beat Minnesota. It is that a 132–1 law lasted two months before a court decided it might be unconstitutional. Nobody in a statehouse should feel good about that — and nobody at xAI should feel good about why it happened.

And state-level tech lawmaking is in flux everywhere — Newsom just vetoed America's first smart-glasses privacy law.

Sources

Reuters reporting by Mike Scarcella on the 8th Circuit's 2 October injunction order and the underlying case (via Channel News Asia, WNCY and Northland News Radio, 2–3 October 2026); Reuters on xAI's September appeal filing, including Judge Donovan Frank's 4 September order and xAI counsel Robert Dunn's background (14 September 2026); BigGo Finance on the district court proceedings, HF 1606's passage (House 132–1, Senate 65–0; Governor Tim Walz's 7 May signing), Grok usage figures and xAI's account-suspension and NCMEC-reporting numbers; RecordingLaw on the district court docket timeline (TRO denial 31 July, preliminary injunction denial 4 September, injunction-pending-appeal denial 9 September, motion to dismiss stayed 1 September); Startup Fortune on the law's provisions; Cryptopolitan and Brandicon Image on Attorney General Keith Ellison's statements and the law's legislative history. The 3-million-images figure is as reported in court-proceedings coverage; xAI's 52,222 suspended accounts and 73,604 NCMEC reports are the company's own reported figures.

Daniel Rao covers AI safety, frontier labs and technology policy for TechPulse USA. He writes about the people and institutions deciding what artificial intelligence is allowed to do.