Minnesota's Nudification Ban Passed 132 to 1. xAI Says It Also Bans Swimsuits.

In 2024, Molly Kelley found out that a close family friend had been feeding photographs of her into a nudification website. He had done the same to roughly 80 other women, all of them pulled from public social media accounts.

Then she found out that almost nothing about it was illegal. The fabricated images sat on his computer and had not been distributed, which put them outside Minnesota's revenge-porn statute. None of the women was a minor. Kelley started calling anyone who would pick up, eventually reached State Senator Erin Maye Quade, and spent the next two years pushing a bill.

That bill is HF 1606. It takes effect on Saturday, August 1.

On Monday, July 27, lawyers for xAI walked into the U.S. District Court for the District of Minnesota and asked a federal judge to stop it. The case is docketed as 0:26-cv-03425. The company is seeking a preliminary injunction against Attorney General Keith Ellison before the statute becomes enforceable, and it wants the law declared unconstitutional outright.

Governor Tim Walz's response arrived on X shortly after the filing became public: "See you in court, creep."

The insult is easy. The legal question underneath it is not, and it will not stay in Minnesota. HF 1606 is the first American statute to attach a per-image dollar penalty to the company that shipped the model rather than the person who typed the prompt. Whichever way the District of Minnesota rules, every state legislature drafting one of these bills will rewrite it afterward.

The Vote Was Not Close and the Law Has No Safe Harbor

Minnesota's legislature is split evenly between the parties, an arrangement that normally guarantees nothing passes. HF 1606 cleared the House 132 to 1 and the Senate 65 to 0. Walz signed it earlier this year. Representative Jess Hanson, DFL-Burnsville, co-led it in the House; Maye Quade, DFL-Apple Valley, carried it in the Senate.

The statute prohibits anyone who owns or controls a website, application, software or other digital service from letting users access or use that service to nudify an image: to alter or generate an image or video depicting an intimate part of an identifiable person that was not in the original source material. Both individual victims and the attorney general's office can sue, and the exposure is $500,000 per unlawful incident.

Targeting the platform instead of the user was deliberate. Minnesota's authors concluded that chasing individual creators after an image has already moved through closed messaging apps and school group chats does not stop the harm at its source.

The federal Take It Down Act, signed in 2025, obliges platforms to remove non-consensual intimate imagery within 48 hours of a victim's request. That is a distribution remedy. It arrives after the image exists.

Hanson did not treat the lawsuit as a close call.

xAI Concedes the Goal and Attacks the Arithmetic

The complaint opens by giving ground. xAI "does not contest Minnesota's interest in prohibiting the dissemination of artificially generated nude images of real people without their consent." Its objection is scope: the statute "extends far beyond that goal" and exposes "a wide array of protected speech to civil liability and government sanctions." The filing calls HF 1606 "an overbroad, content-based ban on free speech and the tools of visual expression."

The sharper argument is about liability structure, not speech. HF 1606 contains no scienter requirement, meaning a platform's exposure turns entirely on what a user does, not on what the operator knew, intended or could have foreseen. There is no safe harbor for an operator whose terms of service ban nudification. There is none for an operator whose filters are, in the filing's phrase, "state-of-the-art and function to near-perfection."

Then the complaint does the multiplication. A platform whose users produced 100,000 prohibited images would face $50 billion in potential liability, even if the same platform had blocked every prior attempt. The arithmetic is not in dispute: the penalty is half a million dollars per incident, and a hundred thousand incidents is a hundred thousand incidents.

The doctrinal frame is familiar from platform-liability cases. A content-based restriction triggers strict scrutiny under Reed v. Town of Gilbert, so the fight is entirely about narrow tailoring. Underneath sits Smith v. California, the 1959 ruling that a bookseller held strictly liable for every book on the shelf will stop stocking books nobody has read. Swap in an image model and the collateral-censorship argument writes itself.

The Complaint's Best Example Is a Photo the President Posted

For overbreadth, xAI needed an image nobody would call nudification. It found one on Donald Trump's own account.

The complaint points to an AI-generated picture the president posted showing himself, the vice president, several cabinet members and an unidentified woman in bathing suits in the Lincoln Memorial Reflecting Pool. Because HF 1606 does not narrow what counts as an "intimate part," the filing argues, the bare chests in that photograph are covered.

Shirtless men. People in shorts. Swimsuit photos. Political satire, medical illustration, art.

There is a technical version of the same point that engineers will recognize faster than lawyers will. A modern image editor has no separate nudification module to switch off. The inpainting pipeline that puts someone in a different jacket is the pipeline that removes one, so enforcement lives entirely at the prompt-filtering and output-monitoring layers, and no classifier reaches a zero false-negative rate across every phrasing a determined user will try. A statute priced per bypass asks for a guarantee the architecture cannot give.

Grok's Own Record Is the State's Strongest Answer

xAI arrives at this argument carrying a specific history, and it is the reason the case looks different from the company's earlier fights with California and Colorado over disclosure and algorithmic-discrimination rules.

In the eleven days between December 29, 2025 and January 8, 2026, immediately after Musk promoted Grok's image-editing feature on X, researchers at the Center for Countering Digital Hate estimated the tool generated roughly 3 million sexualized images, including about 23,000 that appeared to depict children. On January 14, X said it would put technological measures in place to stop Grok's account from editing images of real people into revealing clothing, and moved photo editing behind a paid subscription.

On March 16, 2026, Lieff Cabraser Heimann & Bernstein and Baehr-Jones Law filed a class action in the Northern District of California on behalf of three victims, alleging Grok was used to generate child sexual abuse material from their real photographs and that xAI profited from the capability while declining to adopt standard prevention measures. xAI disputes the allegations. Teenagers in Tennessee brought a separate suit earlier this year, which we covered in Tennessee Teens Sue xAI, Alleging Grok Generated Sexualized Images From Their Photos.

xAI's counter-record, drawn from its own filings, is more than 50,000 accounts suspended, more than 70,000 reports to the National Center for Missing & Exploited Children, and 244 arrests. Those figures are the company's and are not independently verified. They also measure volume: 70,000 reports is not the profile of a tool nobody misuses.

Minnesota's Position Is That There Is Nothing Here to Debate

Ellison had not been served when the news broke. He responded anyway.

The state's substantive answer is that xAI is litigating hypotheticals. No prosecutor is going to bring an action over cabinet officials in swim trunks, and courts routinely construe statutes narrowly rather than striking them when a reasonable reading avoids absurd results. The ACLU of Minnesota, which does not usually line up behind content-based restrictions, backed HF 1606.

The most interesting objection to the law comes from the opposite direction, and it is not about Grok at all. xAI can afford to geofence a feature and litigate for two years. A four-maintainer open-source image model cannot, and neither can a small photo-editing startup that ships background removal and discovers it has acquired uncapped liability. If overbreadth chills anyone, it chills them first.

Washington is pushing from the same direction with different motives. The Trump administration said last year it would challenge state-level AI laws, and this year published what the White House called a comprehensive national legislative framework, arguing that "a patchwork of conflicting state laws would undermine American innovation and our ability to lead the global AI race." That framework has been rewritten more than once, as we tracked when Trump scrapped a finished AI order at the last minute. A federal preemption campaign and a private First Amendment challenge now aim at the same target from opposite ends of the building.

This is xAI's third state challenge in seven months, after suits against California's training-data disclosure law and the Colorado AI Act. Minnesota's is the one with a victim's name in the legislative history.

Illinois offers the contrast. It wrote its AI rules with industry at the table and got Anthropic's public endorsement, a dynamic we examined in Anthropic Backed the Rules. Illinois Just Made Them Law. Minnesota wrote its rule 132 to 1 and got sued five days before it took effect.

What Changes for Teams Shipping Image Models

• Per-output penalties break the usual compliance model. AI risk work assumes fines scale with corporate conduct. A statute priced per generation makes your throughput your liability.
• Scienter is the clause to read first. Whether a law requires knowledge or intent decides whether your filters are a defense or just an expense.
• Undefined statutory terms become blunt refusals. If "intimate part" has no boundary in law, no classifier can be tuned to it, and the safe engineering response is to over-refuse. Geofencing the state is cheaper still.
• Copycat bills are already drafted. Statutes modeled on HF 1606 were pending in California, New York and elsewhere when this suit was filed. The ruling here becomes the template either way.

The Bottom Line

Two things are true, and the case turns on which one a federal judge weighs more heavily.

Nudification tools cause documented, severe harm, overwhelmingly to women and girls, and Minnesota's legislature said so with the closest thing to unanimity a tied chamber can produce. A law that defines "intimate part" loosely enough to reach a swimsuit photo the president published is also, plainly, a law with a drafting problem.

Both of those facts point at the same absence. Nobody has yet written a nudification statute narrow enough to survive strict scrutiny and broad enough to actually stop the conduct. Minnesota went first, so Minnesota gets to find out where the line sits, on a clock that expires Saturday.

Ellison says this is not one of the worthy debates about AI policy. The filing on his desk says the debate was never about nudification. It is about who pays when a general-purpose model does something specific, and Molly Kelley spent two years demonstrating that the answer, until now, was nobody.

Sources

• xAI v. Ellison, Case 0:26-cv-03425, complaint (D. Minn.) (filed Jul 27, 2026)
• House File 1606, 94th Minnesota Legislature (bill text) (Minnesota Revisor of Statutes)
• Elon Musk's xAI sues to stop Minnesota law banning nudification technology (CBS Minnesota, Jul 28, 2026)
• Elon Musk's AI company sues Minnesota Attorney General Ellison over nudification ban (Minnesota Reformer, Jul 28, 2026)
• SpaceX's xAI sues Minnesota over law to ban 'nudify' apps (CNBC, Jul 28, 2026)
• Musk's xAI sues Minn. over first-in-the-nation law banning 'nudification' technology (ABC News, Jul 28, 2026)
• Minnesota passes the nation's first ban on 'nudification' apps (The 19th, Apr 2026)
• xAI seeks preliminary injunction against enforcement of Minnesota ban on 'nudification' (MLex, Jul 2026)
• xAI Challenges Minnesota Nudification Law With No Safe Harbor, No Scienter (Tech Times, Jul 29, 2026)
• Grok floods X with sexualized images of women and children (Center for Countering Digital Hate, Jan 2026)
• LCHB files class action on behalf of minor victims alleging xAI's Grok generated AI sexual exploitation images (Lieff Cabraser, Mar 16, 2026)
• ACLU of Minnesota on HF 1606 (ACLU of Minnesota)
• Musk's xAI, SpaceX combo is the biggest merger of all time, valued at $1.25 trillion (CNBC, Feb 3, 2026)
• Trump AI executive order and its impact on Minnesota regulations (CBS Minnesota)