On August 1, 2026, a Minnesota statute took effect that most AI teams have not read, because it does not carry “AI” in its title. Minn. Stat. § 325E.91 bans “nudification technology,” and it is one of the most aggressive pieces of AI-adjacent law any U.S. state has passed. It reaches the people who run the tool, not just the people who misuse it, and the penalties are large enough to change how a product gets built.
What the law prohibits
A person who owns or controls a service must not allow a user to access, download, or use it to “nudify” an image, and must not nudify an image on a user’s behalf. Separately, no one may advertise or promote such a service. Two definitions set the boundary:
- “Nudify” is a two-part test. An image is altered or generated to depict an intimate part not shown in an original image of an identifiable person, and the result is so realistic that a reasonable person would believe it belongs to that person. Cartoonish or obviously synthetic output falls outside.
- “Identifiable” is a low bar. The person can be identifiable from the image by themselves or by a single other person. It does not take public recognizability.
Why it is sharper than it looks
- The penalty is per incident. The Attorney General may seek a civil penalty of up to $500,000 for each unlawful access, download, or use. On a service with any volume, exposure scales with usage, not a single headline fine.
- Private right of action. A person depicted can sue for compensatory damages including up to three times actual damages, punitive damages, and attorney fees. Plaintiffs, not just the state, will drive enforcement.
- No intent element. The operative verb is “allow.” On its face this reads as strict liability, which is the exact basis of the constitutional challenge now pending.
- Jurisdiction runs on the person depicted. A Minnesota court has jurisdiction if the plaintiff or defendant resides in the state. A company with no Minnesota users, servers, or presence is exposed the moment a Minnesota resident is the person in the image.
The one carve-out, and its limits
The prohibitions do not apply where the service requires the technical skill of a user to produce the result. That is a general-purpose-tool exemption: a professional editing suite demanding substantial human skill and judgment sits outside, while a one-click app sits inside. The line is drawn at how much human skill the tool requires, not at how it is marketed. A general-purpose generative model with a low-friction path to the prohibited output is not obviously protected by it.
The statute also says it does not alter Section 230 and must be read consistently with federal law. Read that as a drafting hedge, not a grant of immunity. Whether Section 230 protects a provider who supplies the generating tool, rather than hosts a third party’s content, is precisely the contested question. Do not assume it covers you.
What to do now
1. Find the path to the output. If any surface you own or host can produce a realistic nude of a real person with low friction, treat it as in scope, regardless of intent. 2. Raise the friction or close the path. Input and output filtering, prompt and image restrictions, and reliable refusal behavior are the controls that matter here. 3. Do not lean on Section 230 or “we didn’t mean to.” Neither is a defense on the face of this statute. 4. Map your U.S. consumer image and video surfaces by state. Minnesota is first, and its damages structure invites plaintiff-side filings.
An honest limitation
We are describing an enacted Minnesota statute, in force since August 1, 2026 and drawn from the Office of the Revisor of Statutes text. The constitutional challenge to it is real and ongoing, and reported outcomes — including a denied request to temporarily block the law — come from news coverage rather than the court docket, so treat the litigation as unsettled. What is not unsettled is that the law is in force, and a denied injunction does not change a present obligation. Verify the current text and any ruling before relying on a specific reading.
This briefing is general information from Sentinel Assurance Group, not legal advice. Regulatory dates and requirements change — we maintain these briefings, but verify against primary sources and counsel before acting. Last reviewed August 10, 2026.
See how a Gap Assessment maps your exposure →Running image or video AI?
Let’s find the path to the output.
The free AI Risk Exposure call maps your consumer AI surfaces to the strict-liability rules now emerging, and shows where a user could reach output you never intended.
Book the call →