Minnesota's anti-nudification law survives xAI's challenge — the precedent for image models

Elon Musk's xAI wanted Minnesota's AI "nudification" ban frozen while it fights the measure in federal court. A judge said no — twice. The law, the first in the nation to hold AI image providers strictly liable for generating fake nudes, is now in force while the constitutional fight grinds on. And that fact alone is the story: even before any court has decided whether the law is constitutional, state-level regulation is already reshaping what an AI product can do.

What happened, in order#

Minnesota's AI nudification ban — HF 1606 — was signed by Gov. Tim Walz on May 7, 2026, after passing the state House 132–1 and the Senate 65–0. In combined votes, 197 lawmakers supported it; one opposed it. The law was scheduled to take effect August 1.

xAI, developer of the Grok chatbot and its Grok Imagine image tool, sued the state on July 27, 2026 — nearly three months after the law was signed and days before it took effect. The company argued the statute violates the First Amendment and exposes a wide range of protected speech to liability.

On July 29, xAI asked the court for a temporary restraining order halting enforcement. Judge Donovan W. Frank of the U.S. District Court for the District of Minnesota denied it on July 31, one day before the law took effect.

xAI then sought a preliminary injunction, which Minnesota Attorney General Keith Ellison's office opposed on August 18. On September 4, Frank denied that too. The same day, xAI filed notice that it would ask the U.S. Court of Appeals for the Eighth Circuit to review the denial.

What the law actually does#

The statute prohibits companies that own AI tools from letting users generate realistic images or videos that add intimate body parts to photos of identifiable people — and from producing such images on a user's behalf. The Minnesota attorney general can seek civil penalties of up to $500,000 for each violation, plus injunctive relief. Individuals depicted in the images can also pursue their own claims.

Two features make Minnesota's approach different from every other anti-deepfake regime:

1. It targets the provider, not just the user. Traditional approaches — including the federal Take It Down Act — punish the person who creates or distributes the nonconsensual image. Minnesota flips the frame: the company offering the tool is liable if its product is used to "nudify" someone, regardless of who pressed the button.

2. It's strict liability. Unlike the federal Take It Down Act and a 2025 Texas deepfake law, which hinge on whether the creator knew the image was nonconsensual, Minnesota's statute removes the knowledge requirement. A provider can be held liable even without awareness that a nude image was generated. That provision is the heart of xAI's constitutional challenge.

The law's bipartisan momentum came partly from reports of a man who used social-media photos to fabricate sexual images of more than 80 women he knew. In its September 4 press release, Ellison's office also cited a report estimating that Grok generated approximately 3 million sexualized images — including about 23,000 that appear to depict children — in just 11 days after a new image-editing feature launched. xAI has countered with its own enforcement numbers: the company says it suspended 52,222 accounts this year and filed 73,604 reports with the National Center for Missing and Exploited Children.

Why the judge said no — twice#

Frank's ruling was narrow but instructive. It did not decide whether the law violates the First Amendment; the constitutional challenge continues. Instead, the judge held that xAI had not met the standard for emergency relief, for several reasons:

• The delay hurt. xAI waited nearly three months after the statute was signed to sue, then sought emergency relief three days before the law took effect. Frank wrote: "If xAI genuinely feared irreparable harm, the Court does not doubt it would have acted more quickly to file this action and move for injunctive relief. The fact that it did not do so strongly suggests a lack of irreparable harm."
• Money isn't irreparable harm. xAI warned of penalties up to $500,000 per violation and claimed compliance changes could cost money and users. The court found financial harms don't qualify as irreparable because they could be recouped if the law is ultimately struck down — and that commercial injuries like user loss from increased moderation didn't demonstrate imminent harm either.
• xAI already changed the product. By the time the court considered the injunction, xAI said it had disabled nudification capabilities for Minnesota users and increased moderation of image-editing prompts from Minnesota users by more than 10 percent. That compliance effort undercut the claim that the company was facing imminent catastrophe.
• The public interest tipped steeply toward the state. Frank noted that Minnesota had "enacted, democratically and nearly unanimously, a civil statute aimed at protecting the public from the undisputed harm stemming from AI-nudification technology" — and that "it is evident, and xAI does not dispute, that the State has an interest in curbing the generation of those images."

xAI's case against the law#

The 38-page complaint, filed July 27, makes three broad arguments:

1. Overbreadth. xAI contends the statute can restrict constitutionally protected expression — including images created with the subject's consent or created by the person depicted themselves — and that its definition of "intimate part" is so broad it covers body parts routinely shown in public.
2. No safe harbor. The company argues the law provides no protection for providers making good-faith efforts to prevent nonconsensual nudification, instead holding them responsible for user-generated images even when safeguards are in place.
3. Self-regulation suffices. xAI says its policies already prohibit "any illegal, harmful, or abusive activities" that breach privacy, including nudifying real people or sexualizing children, and that it suspends or terminates accounts that violate those terms.

Ellison's office counters that company policies are beside the point: voluntary moderation can be loosened at any time, whereas the law imposes government enforcement and financial penalties. The attorney general has called the ability to produce AI nudes without consent "appalling" and framed the law as protecting "the dignity of the people of Minnesota."

What it signals for image-model liability#

The First Amendment question is still open, and the Eighth Circuit appeal will be the next inflection point. But the rulings so far already carry three signals for every company building image generation:

Liability is moving from misuse to the platform. The dominant industry defense — that AI companies are responsible for product safety, not for what users do with the product — is exactly what Minnesota rejected. The state treats nudification tools like inherently hazardous products: if you ship one, you're liable for what it does, knowledge or no. Whether that theory survives First Amendment scrutiny will shape how much safety engineering every image lab needs to do up front.

Geographic fragmentation is already here. xAI's complaint acknowledged the alternative it proposed: suspend the relevant features for Minnesota users rather than risk fines. The company has already built a state-specific version of Grok. Other image generators — from OpenAI, Google, Adobe, Midjourney, and open-weight models behind APIs — will face the same calculus if more states follow Minnesota's lead. One product, fifty rulebooks.

Laches applies to billionaires too. Frank's ruling is a procedural lesson for the AI industry's legal teams: if you believe a law poses an existential compliance threat, sue early. Waiting until the eve of enforcement doesn't just look bad — it becomes evidence that the harm isn't irreparable.

For builders of image models, the practical takeaway is blunt: consent-based safety filters and terms-of-service language may no longer be enough. If strict provider liability spreads, "the user misused it" stops being a defense — and preventing nonconsensual sexual imagery becomes a design requirement, enforced with $500,000-per-violation stakes.

The takeaway#

Minnesota's law survived its first two court tests not because it was found constitutional — that question remains open — but because xAI couldn't show it was being irreparably harmed in the meantime. Yet the outcome is the same in practical terms: a strict-liability regime for image-model providers is now operating law in one state, Grok already behaves differently inside Minnesota's borders, and the Eighth Circuit will decide whether that experiment continues. Every frontier lab shipping image generation is now building under a new legal theory — that the tool itself, not just its misuse, creates the liability.