No Fault, No Exceptions

In May 2026, Minnesota passed a law aimed at one of the ugliest uses of generative AI: undressing real people without their consent. The statute, H.F. 1606, is now Minn. Stat. § 325E.91 under the title “Prohibition on Nudification Technology.” Lawfare described the vote as nearly unanimous. Governor Tim Walz signed it on May 7, and it took effect on August 1.

Five days before that date, xAI sued. Elon Musk’s company, now captioned as SpaceXAI LLC, argues that the law violates the First Amendment. On September 4, a federal district judge in Minnesota refused to block the law. On October 2, the Eighth Circuit granted xAI an injunction pending appeal in a single sentence with no reasons. The appeal is now being briefed.

I want to be careful here, because this case invites two lazy readings. The first is that Musk’s company is fighting for the right to make deepfake porn. The second is that any law against nudify apps is censorship. Neither is right. The real question is narrower and more interesting: can a state hold a company liable for every nudified image its users make, regardless of fault, intent or the safeguards it built?

What the law says

The statute does not punish the person who makes or shares an image. It targets the tool. Anyone who owns or controls a website, app, program or service may not “allow a user to access, download, or use” it “to nudify an image or video,” may not “nudify an image or video on behalf of a user,” and may not advertise such a service.

To “nudify” means altering or generating an image or video “to depict an intimate part not depicted in an original unaltered image or video of an identifiable individual,” where the result is “so realistic that a reasonable person would believe that the intimate part belongs to the identifiable individual.” There is one exemption: services that require “the technical skill of a user,” defined as the substantial application of individualized technological or artistic skill and judgment by a human creator. In practice, that means Photoshop is fine and one-click undressing is not.

Enforcement runs on two tracks. A depicted person can sue for damages, punitive damages and an injunction. The Attorney General can seek a civil penalty of up to $500,000 “for each unlawful access, download, or use.” The statute does not mention intent, knowledge, negligence, consent, or whether the image was ever shared. That is what makes it a strict-liability law, and why it attracted a well-funded challenger.

What xAI argues

According to The Verge’s reading of the complaint, xAI concedes that the state has a valid interest in stopping the nonconsensual spread of fake nudes. Its argument is that Minnesota went far beyond that. The law, it says, reaches consensual images, images of adults that are suggestive rather than abusive, and images with humorous or artistic value. It also argues that liability attaches “even if the company has deployed near-perfect, state-of-the-art technical controls.” Bloomberg Law summarized the core claim: this is a content-based restriction on speech, presumptively unconstitutional unless narrowly tailored to a compelling interest by the least restrictive means.

xAI also points to laws that, in its view, already do the job in a constitutional way: Minnesota’s own rules against disseminating sexual deepfakes and the federal Take It Down Act. Those laws target the publication of nonconsensual intimate images, or NCII. The nudification statute instead targets the capability.

The company is a hard plaintiff to sympathize with. The Verge recounts that in January 2026, Grok produced a flood of sexualized edits of real people. The Center for Countering Digital Hate estimated that Grok generated about 3 million sexualized images in the 11 days from December 29 to January 8, around 23,000 of which appeared to depict children. Sarah Jeong put it sharply at The Verge: if the “good” laws were working, January would not have happened. That is a fair point, and it explains why legislators reached for a blunter tool.

The doctrine Minnesota has to get past

Constitutional law does not ask whether the plaintiff is likable. It asks what the law does. Three lines of precedent matter here.

The first is about content. Under Reed v. Town of Gilbert (2015), a law that singles out speech because of its subject matter or message is content-based and faces strict scrutiny. Images are speech, and AI-generated images are still images. Minnesota reportedly argues that the statute regulates technology, not expression, and should get the gentler intermediate scrutiny. xAI says that a law defined by what an image depicts is content-based by definition. Which test applies will probably decide the case.

The second is about the categories of unprotected speech. Child sexual abuse material falls outside the First Amendment under New York v. Ferber (1982), and obscenity has long been unprotected. But in Ashcroft v. Free Speech Coalition (2002), the Supreme Court struck down a ban on computer-generated sexual images that involved no real child and were not obscene. The Court refused to expand the unprotected categories just because the images were disturbing. Minnesota’s law is broader than child abuse material or obscenity. It covers any realistic intimate depiction of an identifiable adult, including one that adult asked for.

The third line is the one most relevant to the phrase “strict liability.” In Smith v. California (1959), the Supreme Court struck down a Los Angeles ordinance that made booksellers liable for possessing obscene books even if they did not know what was in them. The Court reasoned that a seller who could be punished without knowledge would stock only what they had personally inspected, and the public’s access to lawful books would shrink. More recently, in Counterman v. Colorado (2023), the Court required at least recklessness before someone can be punished for true threats, in part to avoid chilling protected speech. Neither case is about AI. Both reflect the same principle: when a law burdens speech, the Constitution is suspicious of liability without a culpable mental state, because the predictable result is over-caution.

Over-caution is exactly what xAI says it faces: “no practical choice but to restrict Grok Imagine’s image-editing features in various ways,” as the complaint puts it according to The Verge. To some readers, that sounds like the law working. To a court, it may sound like a chilling effect.

Where the case stands

It helps to be precise, because the coverage often is not. Recording Law’s docket summary lays out the procedure:

Judge Donovan W. Frank denied a preliminary injunction on September 4. He relied on xAI’s delay in suing and on the weakness of its irreparable-harm showing, not on the First Amendment.
The Eighth Circuit’s October 2 order reads, in full: “Appellant’s motion for injunction pending appeal is granted.” It bars the Attorney General from enforcing the law against xAI while the appeal proceeds.
The order does not touch the private right of action. It does not invalidate the statute for anyone else, and it says nothing about the merits.
xAI’s opening brief was filed on October 30. The Attorney General’s brief is due November 30.

The groups that filed amicus briefs supporting xAI’s motion are worth noticing: the Foundation for Individual Rights and Expression, the First Amendment Lawyers Association and the Woodhull Freedom Foundation. Woodhull is one of the country’s most prominent sexual-freedom organizations. Its presence makes clear that this is not only a tech-industry fight. Sex-positive civil libertarians also have a stake in how far a state can go in policing realistic depictions of bodies.

What a better law would look like

I do not think the answer is to protect nudify apps. A service built and marketed to undress strangers is a harassment machine. Victims of deepfake porn describe harms that are real and lasting. Some of those victims are children, as cases in Pennsylvania and elsewhere keep showing — including the reporting gap that the state’s new mandated-reporter law tries to close. A state is right to go after the business model.

But the Minnesota statute does not target the business model. It targets the result, no matter how the result came about. A general-purpose image tool that blocks nudification and gets jailbroken by one determined user faces the same $500,000 per use as an app that advertises undressing anyone from a single photo. A couple who make an explicit image of themselves together, or an artist who paints a consenting model in a more realistic style, fall within the same definition. The law has no consent exception, no mens rea, and no safe harbor for reasonable safeguards.

There are narrower options, and nothing in First Amendment law forbids a state from using them. Liability could turn on knowledge or recklessness. It could cover services designed or marketed for nonconsensual nudification, or providers that fail to adopt reasonable safeguards. Consent could be an explicit defense. Penalties could scale with fault. Each of these choices would make the law easier to defend and harder for an xAI to attack, and each would keep the focus on the people who actually do the harm.

Desire is not the enemy here. Consensual nudity, explicit art and sexual play with AI are part of adult life, and a sex-positive approach to technology should defend them. Nonconsent is the enemy, and the law works best when it says so precisely. Minnesota’s statute names the right problem. Whether its no-fault, no-exceptions design can survive the First Amendment is now up to the Eighth Circuit, and its answer will shape how every other state drafts the next nudify law.