On 2 October 2026 the Eighth Circuit granted xAI an injunction pending appeal, barring Minnesota Attorney General Keith Ellison from enforcing the state's nudification ban against the company. The order gave no reasoning, and it does not decide whether the law is constitutional. It reverses the practical effect of U.S. District Judge Donovan Frank's early-September ruling, which found xAI had not shown irreparable harm and had waited too long to seek emergency relief. The merits are still to be argued, and xAI's opening appellate brief was due 29 October.
The strongest case for Minnesota
The state's case deserves a fair hearing. Minnesota told the court the law targets the "staggering amount" of child sexual abuse material that AI products such as Grok Imagine generate. Victims of fake nude images suffer real and often lasting harm. Takedown after the fact is a poor remedy when a tool can produce a new image in seconds. A legislature that wants to act upstream, at the tool, is making a defensible policy judgment, and the vote shows how widely it was shared: House File 1606 passed the House 132-1 and the Senate 65-0.
What the statute actually does
The law, Chapter 72 of the 2026 session laws, adds section 325E.91 to Minnesota Statutes. It was signed on 7 May 2026 and took effect on 1 August. It bars anyone who owns or controls a website, app or other service from allowing a user to "access, download, or use" it to nudify an image or video. It also bars nudifying an image on a user's behalf, and it bars advertising such a service.
"Nudify" means altering or generating an image so realistically that a reasonable person would believe the intimate part belongs to an identifiable individual. The attorney general can seek civil penalties of up to $500,000 for each unlawful access, download or use. Depicted individuals get a private right of action for up to three times their actual damages, plus punitive damages and fees. The law leaves 47 U.S.C. § 230 untouched.
Two drafting choices matter for the First Amendment challenge. First, the penalty attaches to each access or use, not to each harmful image distributed. A general-purpose image model with millions of users could accumulate exposure far out of proportion to any actual victim count. Second, the statute exempts services that require "the technical skill of a user" to nudify an image. That carve-out sorts tools by how easy they are to use, not by the harm they cause. A court asking whether the law is narrowly tailored will notice that a skilled user of an open-source model produces the same image and falls outside the ban.
Why the injunction does not settle much
It is tempting to read the Eighth Circuit's order as a verdict on the merits. It is not. An injunction pending appeal preserves the status quo while a panel considers the case. The order gave no reasoning, so nobody outside the court can say whether the judges weighed the speech question, the delay finding or the balance of harms. Treat any commentary that claims the court found the law likely unconstitutional as speculation. The order, as reported, also protects xAI specifically. The statute remains on the books for every other covered service.
That unevenness is a cost. A law that binds some tool makers and not others during a long appeal creates a compliance map nobody can plan around, and it hands one defendant a competitive advantage in a market where the harm is the product's misuse.
A narrower federal tool already exists
Congress has already legislated in this space, and it chose a different design. The TAKE IT DOWN Act's notice-and-removal rules became enforceable on 19 May 2026. The FTC says a covered platform that receives a valid request "must remove the content—along with any known identical copies—within 48 hours." The Act reaches "digital forgeries," including images created or altered with AI. Platforms that violate it face potential civil penalties of $53,088 per violation.
The federal model attaches liability to a concrete event: a victim's request that the platform ignores. That is proportionate and administrable, and it leaves the design of image-generation tools to ordinary product-safety and tort law. It is not a complete answer, because takedown is reactive. But it is a precedent for regulating the harm without regulating the capability itself, and it is a model that courts have less reason to doubt on speech grounds.
What a better Minnesota law would look like
The goal of preventing synthetic sexual imagery of real people is legitimate, and pro-innovation does not mean indifferent to victims. The better drafting path is available:
- Tie penalties to the generation or distribution of an actual depicting image of an identifiable person, not to every access or use of a tool.
- Replace the "technical skill" exemption with a duty-of-care standard: reasonable safeguards against producing sexualised images of real people, judged by effectiveness.
- Keep the strong private remedy for victims, which targets actual harm and needs no state enforcement discretion.
- Add explicit carve-outs for journalism, satire and clearly fictional or consenting-adult content, which xAI has argued the statute lacks.
States are right to be impatient. But a statute that a court strikes down protects no one, and a statute drafted around a tool's accessibility rather than around the victim's injury invites exactly the litigation Minnesota now faces. When the Eighth Circuit eventually explains itself, legislators elsewhere should read the opinion before copying the text.