US court pauses Minnesota AI nudification ban: what UK and EU providers should do

What the US court did, and did not, decide

Reuters reported that a three-judge Eighth Circuit panel granted xAI's request for an injunction pending appeal on 2 October. The appeal follows the US District Court for the District of Minnesota's decision of 4 September 2026 refusing a preliminary injunction against the state law.

The district court did not decide the constitutional merits. It found that xAI had delayed seeking emergency relief and had not shown irreparable harm. It expressly described the constitutional issues as complex and said they would receive full consideration in the ordinary course. The appellate intervention changes the interim enforcement position, but it does not answer those merits questions.

According to Reuters, xAI argues that Minnesota's law is a content-based restriction contrary to the First Amendment. Minnesota says the law targets the harm caused by non-consensual intimate imagery and child sexual abuse material. Those are the parties' positions, not findings that can yet be treated as settled law.

The Minnesota law under challenge

Minnesota's official legislative summary says the law took effect on 1 August 2026. It prohibits a person controlling a website, application, software or other service from allowing a user to access, download or use technology that can nudify an image or video, or from nudifying content for the user. It also restricts advertising and promotion of that technology.

The state law provides civil remedies and permits a civil penalty of up to $500,000 for each unlawful access, download or use. Those provisions explain the commercial significance of the dispute, but the Eighth Circuit's interim order means Minnesota cannot presently enforce the challenged law against xAI while the appeal proceeds, according to Reuters.

The pause is not a global safe harbour. It concerns one US state's law and one pending appeal. A provider must still assess other US federal and state rules, criminal law, civil claims, platform duties, contractual exposure and every non-US regime into which the service is offered.

UK offences are already in force

In England and Wales, sections 66E and 66F of the Sexual Offences Act 2003 came into force on 6 February 2026. Crown Prosecution Service guidance explains that they criminalise creating, or requesting the creation of, a purported intimate image of an adult without consent and without a reasonable belief in consent. The precise offence and available defences depend on the facts and statutory wording.

A separate tool offence in section 66I came into force on 29 June 2026. It covers making, adapting, supplying or offering to supply a thing where a reasonable person would regard the circumstances as giving rise to an inference that it is made or adapted for use as a purported intimate-image generator. The statutory definition of a thing includes a program, electronic information and a service, so product and API architecture can fall within the analysis.

The maximum sentence for the section 66I offence is three years' imprisonment and/or an unlimited fine on indictment. A person has a defence if they prove that they took all reasonable steps to prevent the thing being used to create non-consensual purported intimate images. This is a fact-sensitive criminal-law defence, not a product certification or a checklist safe harbour.

The Home Office says the tool offence has extraterritorial effect for UK persons and bodies. Businesses should therefore map incorporation, personnel, development, supply and marketing activities rather than treating geoblocking at the user interface as a complete answer. CPS guidance and the Home Office circular explain the legislation, but the statutory provisions remain the binding law.

The EU prohibition applies from 2 December 2026

Regulation (EU) 2026/1744 amended Article 5 of the EU AI Act. From 2 December 2026, a new prohibition addresses placing on the market, putting into service or using an AI system to generate or manipulate realistic intimate or sexually explicit content depicting an identifiable person without that person's freely given, specific, informed, unambiguous and explicit consent.

A related prohibition addresses AI systems intended to generate or manipulate child sexual abuse material or child sexual exploitation performances, subject to the legislation's limited without-right defence under national law. These provisions are part of binding EU legislation, but their application date must not be brought forward: they are not yet applicable on 3 October 2026.

Provider scope is not unlimited. The consolidated Article 5 text links provider liability to the system's intended purpose or to a reasonably foreseeable and reproducible outcome, without significant technical modification, where the system lacks reasonable and adequate technical safety measures and safeguards that reliably prevent and correct misuse. A deployer's prohibited use is assessed separately.

That wording makes evidence central. A provider needs to understand foreseeable workflows, reproduce known bypasses, evaluate whether safeguards actually work and remediate failures. A policy statement or prompt refusal, without testing and enforcement across image upload, editing, API and fine-tuning routes, is unlikely to answer the whole statutory test. The last point is a compliance inference from the legislation, not an official finding about any particular product.

Practical actions for AI founders and technology providers

1. Map the product and the actor. Record the intended purpose of each model, endpoint, editing mode and third-party integration. Separate provider, deployer, platform and user roles for each jurisdiction rather than applying one global label.
2. Test foreseeable misuse. Red-team prompts, image-to-image editing, face insertion, fine-tunes, API chaining, obfuscation and repeated low-volume attempts. Preserve the model version, configuration, test inputs, outputs, bypass route and remediation evidence.
3. Build consent around the depicted person. Where intimate or sexually explicit content could involve an identifiable adult, design a process capable of establishing that person's specific and explicit consent for the relevant generation or manipulation. An uploader's acceptance of terms does not by itself prove another depicted person's consent.
4. Use layered technical safeguards. Combine input and output detection, identity and age protections, restricted features, abuse-rate controls, account and device signals, human escalation and action against repeat evasion. Measure false negatives and bypass resistance, not only ordinary prompt refusals.
5. Create a release gate. Require legal and safety approval before enabling new image-editing functions, model weights, fine-tunes or looser safety settings. Define stop criteria, rollback authority and an emergency route to disable a risky feature or endpoint.
6. Operate a rapid reporting and response process. Give affected people and trusted reporters a clear route to flag content, preserve evidence lawfully, remove or restrict material where required, investigate linked accounts and prevent reappearance where applicable law permits and requires it.
7. Document the UK reasonable-steps case. Keep dated risk assessments, tests, safeguards, incident decisions, supplier controls, marketing restrictions, staff training and remediation records. The evidence should show what was reasonable for the actual product and risk at the relevant time.
8. Control claims and contracts. Avoid marketing a service as uncensored, capable of undressing people or legally compliant everywhere. Allocate notice, investigation, evidence preservation, model-change, audit, takedown and cooperation duties across suppliers, customers and platforms.
9. Maintain a jurisdiction and effective-date register. The Minnesota appeal, current UK offences and 2 December EU application date should be tracked separately. Reassess controls when the Eighth Circuit issues its reasons or final judgment and when official EU guidance or enforcement practice develops.

The compliance conclusion

The Minnesota order is important because it tests how a broad product restriction interacts with US constitutional speech protections. It is still an interim procedural development. Treating it as a merits decision, a repeal or permission to deploy the same product elsewhere would overstate the law.

For UK and EU businesses, the more useful response is to build a defensible control record now. The UK tool offence already makes reasonable preventive steps directly relevant. The EU text will soon test intended purpose, foreseeable and reproducible misuse, and the reliability of safeguards. Product design, testing and evidence therefore matter as much as terms of service.