One statute, one date, one federal complaint

Minnesota's legislature passed the first outright ban in the United States on nudification technology, carried by Representative Jessica Hanson and Senator Erin Maye Quade and enacted as chapter 72. It takes effect on 1 August. The operative sentence is written at the level of the service rather than the user: a person who owns or controls a website, application, software, program or other service must not allow a user to access, download or use it to nudify an image or video, or nudify an image or video on behalf of a user. Advertising or promoting such a service is prohibited on the same terms.

Days before that date, xAI went to federal court against Minnesota Attorney General Keith Ellison, asking for a declaration that the law is unconstitutional and an order barring its enforcement. The exposure it is trying to get ahead of is not symbolic. The statute carries a civil penalty of up to 500,000 dollars for each violation, and it creates a private right of action under which an injured person can seek compensatory damages up to three times actual damages, punitive damages, attorney fees and injunctive relief. Penalties collected are directed to victim services.

What the complaint actually argues

The filing is a speech case, not a technology case. It says the statute imposes an overbroad, content-based ban on free speech and the tools of visual expression, that it extends far beyond that goal, and that it exposes a wide array of protected speech to civil liability and government sanctions. A second line of attack is definitional: xAI contends the law does not adequately define what counts as intimate parts, and the complaint offers examples of images it says are lawful yet would fall inside the prohibition.

The attorney general's response was short and did not engage the constitutional point. Ellison said the office was waiting to be served and would review the lawsuit in court, then added that using AI to generate nude images of people against their will is appalling, that there are plenty of worthy debates to have about AI policy, and that this is not one of them.

Read structurally, the two positions are not really arguing with each other. One side is litigating whether a state may draw a content-based line around a general-purpose image tool. The other is asserting that the conduct the line addresses is not a policy question at all. That gap is why this case will be decided on the first question and will say nothing about the second.

The defence is assembled from parts Europe does not stock

Every load-bearing element of the complaint is American. Overbreadth, content-based restriction, protected expression: these are doctrines of United States constitutional law, and they are the reason a pre-enforcement challenge is even available before a single penalty has been assessed. A European operator watching this docket for a signal about its own exposure is reading a document written in a legal language its own regulator does not use.

The contrast is sharper because the same company is on the other side of a European proceeding about the same product. A member of the United Kingdom parliament, Jess Asato, is pursuing a High Court claim in London against xAI for misuse of private information and breach of data protection after Grok was used to generate sexualised images of her, and the relief sought is an order requiring effective and permanent technical measures so the model cannot produce altered images of her. In that forum the question is not whether the output is expression. It is whether the controller took appropriate measures.

So the same model is the subject of two proceedings eight weeks apart, with the company as plaintiff in one and defendant in the other, and the winning argument in the first is not available in the second. Worth noting for anyone who ships a model into more than one jurisdiction: a federal complaint is a public document, and a description of what your system can generate does not stay in the forum where you filed it.

What to have settled before your own date arrives

If your organisation builds, fine-tunes, hosts or resells a generative image model, the practical lesson is not about Minnesota. It is that the defensive position you can run depends entirely on which regime is asking, and you need to know which ones apply to you before someone else picks the forum. Write down, per market you serve, what the operative duty attaches to: the output, the tool, the deployment, or the data subject in the image.

Then look at what a filing costs you elsewhere. Anything you assert publicly about your model's capabilities is available to every other claimant and regulator with an interest in the same system. Coordinate the positions across jurisdictions before the first one is filed, because after that the argument is fixed and the other forums have already read it.