A federal appeal over Minnesota's new ban on "nudification" technology is emerging as an early test of how far a state can hold an AI developer liable for what users do with its tools. In a Nov. 4 analysis for Lawfare, Jeff Kosseff, a nonresident senior legal fellow at The Future of Free Speech, argues that the Eighth Circuit's eventual opinion in xAI's challenge will likely need to resolve a question courts have not yet answered: what state of mind an AI company must have before it can be penalized for its users' conduct.

The statute

Kosseff describes the law, HF 1606, titled "Prohibition on Nudification Technology," as having passed the Minnesota legislature in the spring by a nearly unanimous vote. It authorizes the state attorney general to seek fines of up to $500,000 from a technology company each time the company allows a user to nudify an image or video, or nudifies one on a user's behalf. The definition reaches realistic altered or generated imagery that depicts an identifiable person's intimate parts not shown in the original.

The design departs from existing state laws on nonconsensual intimate imagery, which target the individuals who create or distribute the material. Under HF 1606, liability attaches to the company whose technology was used. Kosseff reads the statute as imposing strict liability: in his account, a developer remains exposed even if it maintains anti-deepfake policies and technical safeguards and a single user manages to get around them. He also flags the absence of an intent or consent element, the lack of a satire exception, and a broad definition of intimate parts.

The litigation so far

According to the Lawfare piece, xAI, the Grok developer since renamed SpaceXAI, sued Minnesota in late July, days before the law was to take effect, and sought an injunction. The company says it already prohibits users from generating nonconsensual deepfake nudes and that it restricted Grok's image-editing tool in Minnesota in response to the statute. Its complaint contends the law extends well beyond that goal and exposes protected expression to civil liability.

Minnesota's position, as Kosseff summarizes it, is that the statute regulates harmful technology rather than ideas or expression. The state likened the capability to possessing chemical precursors for dangerous drugs, argued that omitting a consent limit is no defect because adults can obtain consensual nude images by other means, and asserted that xAI knows how to identify illegal uses yet has not stopped the harm.

On Sept. 4, U.S. District Judge Donovan Frank denied a preliminary injunction without reaching the First Amendment questions, focusing on xAI's wait of nearly three months after the law was signed before suing; he indicated the constitutional arguments would receive full consideration when he rules on the state's motion to dismiss. xAI appealed directly, with an amicus brief from the Foundation for Individual Rights and Expression, the First Amendment Lawyers Association and the Woodhull Freedom Foundation. On Oct. 2, the Eighth Circuit issued a one-sentence order temporarily barring enforcement of the law against SpaceXAI while it considers the appeal.

The state-of-mind question

Kosseff's central point is doctrinal. He notes that the Minnesota Supreme Court in 2020 upheld the state's individual-user intimate imagery law in part because it required intent and non-consent, elements HF 1606 omits. In his framing, the more rigorous a law's state-of-mind requirement, the more likely it is to survive constitutional review. He distinguishes generative AI from drug precursors on the ground that it can affect a wide range of protected speech, which is why a ruling could reach well beyond nudification. He hedges on the outcome, writing that the Eighth Circuit's decision cannot be predicted with certainty, while treating the early injunction as an indication that the court takes the First Amendment concerns seriously.

Where federal law sits

The case unfolds alongside federal measures aimed at the same harm from other directions. In a January Lawfare article, Renée DiResta and Berin Szóka described the TAKE IT DOWN Act, enacted in May 2025, as criminalizing knowing publication of nonconsensual intimate imagery and requiring platforms to set up notice-and-removal processes by May 2026, with FTC enforcement. In their reading, the act contemplates such imagery hosted on social networks rather than AI providers, so it does not require a provider such as Grok to block users from generating modified versions of an image; they also quoted one lawyer who said notice-and-removal is hardly a remedy because it comes only after the harm has occurred. Minnesota's statute, by contrast, places the obligation on the developer.

Criminal enforcement against individual users is also active. On an Oct. 22 Lawfare Daily episode, Stanford policy fellow Riana Pfefferkorn and University of Edinburgh researcher Cyndie Demeocq discussed 71 federal prosecutions involving AI-generated child sexual abuse material. In 32 of them, prosecutors alleged morphed imagery of real children, a category the episode's host, Renée DiResta, associated with nudifier apps; Pfefferkorn noted that such imagery can come from several kinds of tools. Eight defendants raised First Amendment challenges, and only one succeeded in part. Pfefferkorn argued that prosecutors are managing under existing law and identified victim identification as the larger problem.

That contrast frames the Eighth Circuit's task. In the prosecutions Pfefferkorn and Demeocq reviewed, constitutional challenges to user-focused charges mostly failed; Minnesota's developer-focused, strict-liability model has yet to receive a merits ruling from any court.

Sources