Appeals court freezes Minnesota’s AI ‘nudification’ ban in xAI case

The story

A U.S. appeals court has temporarily blocked Minnesota from enforcing a first-of-its-kind law aimed at artificial-intelligence tools that generate realistic intimate images of identifiable people. The Eighth Circuit granted an injunction sought by xAI while the company appeals a lower-court ruling that had allowed the statute to remain in force. The order changes the law’s immediate status, but it does not decide whether the measure is constitutional.

Minnesota’s law took effect on August 1. It prohibits a person controlling a website, application, software program or other service from letting users employ that product to create a realistic image or video showing an intimate body part that did not appear in the original image of an identifiable person. The statute targets providers rather than only the end user and authorizes state enforcement, including civil penalties of as much as $500,000 for each violation.

xAI sued Minnesota Attorney General Keith Ellison in July, arguing that the law reaches protected expression and violates the First Amendment. The company says the statute is too broad because it lacks exceptions for categories such as artistic, scientific, political, satirical, educational, medical or religious material. It has also argued that providers acting in good faith lack a clear safe harbor. Minnesota counters that the law regulates a harmful technological function and responds to non-consensual sexual imagery, including material involving children.

The procedural history is important. U.S. District Judge Donovan Frank first declined to issue emergency relief and then denied xAI’s request for a preliminary injunction on September 4. His opinion emphasized that xAI waited nearly three months after the law was signed before suing, had not shown irreparable harm and had already deployed geographically targeted controls in Minnesota. The judge said the balance of harms and public interest favored the state while leaving the constitutional issues for fuller review.

xAI appealed in case No. 26-2806 and asked the Eighth Circuit for an injunction while that appeal proceeds. Reuters reported on October 2 that the appellate court granted the request and put enforcement on hold. The court’s action reverses the immediate practical result of the district court’s ruling, but an injunction pending appeal is interim relief. It is not a final judgment invalidating Minnesota Statute § 325E.91.

The case presents a difficult line-drawing problem. Once a convincing synthetic intimate image is generated and distributed, the harm to the depicted person can be rapid and difficult to reverse. Minnesota designed its law to intervene at the product layer before publication. Yet software capable of image transformation can also produce lawful expression, and a provider-facing prohibition may encourage systems to block more content than the state intended if the legal boundary is uncertain and penalties are severe.

The technical question is therefore inseparable from the legal one. A rule that depends on whether a model ‘allows’ a prohibited output raises implementation questions about prompt filters, image matching, age and identity signals, human review, red-team testing and circumvention. No safeguard is perfect. Regulators must decide whether liability turns on the existence of any prohibited output, a provider’s knowledge, its design choices or the reasonableness of controls deployed before and after an incident.

xAI says Grok Imagine includes protections against sexualized depictions of real people and has pursued users accused of bypassing its blockers. Minnesota has cited the scale and severity of abusive synthetic imagery as evidence that provider-level obligations are necessary. Neither position has been finally accepted by a court. The record will need to distinguish claimed safeguards from measured performance and separate deliberate circumvention from ordinary model behavior.

INNOVOX analysis: the injunction makes this dispute an early test of how constitutional speech protections interact with AI product regulation. A binary choice between unrestricted generation and a sweeping platform ban is unlikely to produce a stable policy model. More durable legislation could define prohibited non-consensual conduct precisely, require documented risk controls, establish notice and rapid-removal duties, preserve evidence for victims and provide a conditional safe harbor for providers that meet verifiable standards.

What to watch next is the Eighth Circuit’s treatment of xAI’s likelihood of success and claimed irreparable harm, followed by the merits briefing scheduled in the appeal. Minnesota’s standing arguments and the level of constitutional scrutiny will shape the case. Other states and national regulators will also be watching: a detailed appellate ruling could influence whether future laws target generated content, individual misuse, platform design or some combination of all three.

INNOVOX analysis

The dispute exposes a central design problem in AI regulation: lawmakers want prevention before a non-consensual image spreads, while platform-level bans can also reach lawful or consensual expression. Durable rules will likely need precise definitions, victim-centered remedies, provider duties tied to reasonable safeguards, and narrowly drawn exceptions rather than relying only on either user punishment or blanket product restrictions.

What to watch

Watch for the Eighth Circuit’s explanation of the injunction, xAI’s merits brief, Minnesota’s response on standing and constitutional scrutiny, and whether lawmakers propose a narrower safe harbor for providers that deploy and document effective safeguards.