The EU Nudifier Ban Explained

On December 2, 2026, a new prohibition in the EU’s AI Act starts to apply. It targets AI systems that generate or manipulate realistic sexual or intimate material of an identifiable real person without that person’s consent, and AI systems that generate child sexual abuse material. Politicians and press releases call it the “nudifier ban.” That label is useful, but the legal text is both narrower and more technical than the label suggests. Since this affects anyone who builds, hosts or markets image tools in Europe, including adult-oriented ones, it is worth reading closely before the date arrives.

This is an explainer based on primary texts: the Parliament and Council press releases, Regulation (EU) 2026/1744 as published in the Official Journal, and the consolidated AI Act, including its application-date clause and fines provision. It is my reading, not legal advice. Where I interpret rather than quote, I say so.

How we got here

The ban did not appear in the original AI Act of 2024. It arrived through the “digital omnibus” on AI, a simplification package that the European Commission proposed in November 2025. Most of that package is about deadlines: it postpones the rules for high-risk AI systems to December 2, 2027 for stand-alone systems and August 2, 2028 for those embedded in products. The nudifier ban is part of the same text, and I did not verify at which stage of the legislative process it was added.

Parliament gave its final approval on June 16, 2026, by 423 votes to 57 with 174 abstentions, according to its press release. The Council gave its final green light on June 29. The regulation is dated July 8, 2026, as Regulation (EU) 2026/1744, and the AI Act site run by the Future of Life Institute, which reproduces the Official Journal text, lists July 27, 2026 as its entry into force. The application clause, which I checked in the consolidated AI Act, sets the new prohibition apart from the rest of Chapter II: the older prohibitions have applied since February 2025, but the new points apply only from December 2, 2026.

Parliament’s co-rapporteur Michael McNamara described the result as “an outright ban on AI nudification apps.” The Council’s release put it in plain terms: systems that “generate nude images of real people or edit clothes out in existing photos to reveal intimate parts” are set to be banned from December.

What the text actually prohibits

The ban sits in Article 5 of the AI Act as two new points. Point (ba) prohibits placing on the market, putting into service or using an AI system that generates or manipulates realistic images, video, audio or similar material of an identifiable natural person’s intimate parts, or of an identifiable person engaged in sexually explicit activities, without that person’s freely given, specific, informed, unambiguous and explicit consent. Point (bb) prohibits systems that generate or manipulate child sexual abuse material, within the meaning of the EU’s 2011 directive on child sexual exploitation, except where a “without right” defense applies under national law.

The consent standard is worth pausing on. It borrows the vocabulary of data protection and then adds “explicit.” A person’s silence, a terms-of-service checkbox on someone else’s account, or the fact that a photo is public is not that. For the wider debate on what consent means when a model does the generating, see consent in AI.

Two limiting conditions follow in new paragraphs 1a and 1b, and they decide how the ban works in practice.

For providers, placing a system on the market or putting it into service is prohibited only in two cases. The first is where generating that material is the system’s intended purpose, which describes the classic nudification app. The second is where the system’s design, training, architecture, capabilities or user-facing functions make such output a reasonably foreseeable and reproducible result without significant technical modification, and the system lacks reasonable and adequate safeguards that reliably prevent it, taking foreseeable misuse into account, and that correct observed or reported misuse. A general-purpose image model that can be pushed into producing such images is therefore not banned outright; a general-purpose model that does it easily and has no real safeguards can be.

The recitals say what safeguards might look like: data cleaning, refusal training, safe prompt design, output controls, runtime prompt guardrails, content classification and filtering, usage restrictions, abuse detection, and notice-and-action mechanisms. They add that measures count as adequate if they align with the state of the art and demonstrably prevent or sufficiently reduce the likelihood of the output, including foreseeable circumvention. Recitals are explanatory, not binding in themselves, but this list will shape how regulators read the word “adequate.”

For deployers, use is prohibited only where the deployer uses the system for the purpose of generating or manipulating such material. Under the AI Act’s definition in Article 3(4), a deployer is a person or organization using an AI system under its authority, except where the system is used in the course of a personal non-professional activity. On my reading, the Act therefore reaches the businesses that build and run these tools and the organizations that deploy them professionally, and it does not make a private individual’s one-off experiment an AI Act violation. That does not make private misuse legal. It means other law applies, which I come to below.

Paragraph 1b adds a carve-out for editing that changes nothing about nudity: an AI system that alters material without increasing the exposure of any depicted intimate parts or changing the nature of any depicted sexually explicit activity does not count as manipulation. A background swap or a color correction on an already explicit image the subject consented to would be an example of what that wording appears to allow, but it is my example, not the text’s.

What is not covered

Reading the two points closely shows several things the ban does not do.

It does not ban adult sexual content. The prohibition turns on an identifiable real person and the absence of consent. An erotic image of a fictional, non-identifiable character, or of a person who has given explicit consent, is outside it. AI-generated erotica as such is not what point (ba) targets.
The material must be realistic and the person identifiable. As I read it, cartoon-style output of a fictional figure is not covered by the text. How “identifiable” works when a model produces a face that merely resembles a real person is, in my view, the open question; the text does not answer it.
It is aimed at AI systems, not at every copy of an image. The ban regulates tools and their operators. As I read it, it does not itself create a notice-and-takedown duty for a platform that hosts the output. That is the job of other instruments, which this explainer does not cover.
It does not replace criminal law. The recitals say plainly that the conduct may also violate criminal law and that the prohibitions do not preclude prosecution, and they require member states to respect the principle that nobody is punished twice for the same thing. They also preserve national civil remedies for image, privacy and dignity rights.
Not every use of face-altering technology is caught. A face swap that does not produce intimate or sexually explicit material is outside point (ba). A face swap that puts a real person’s face into explicit material without her consent is inside it.

Who enforces, and how hard

Enforcement is mostly national. Market surveillance authorities in each member state supervise AI systems, and the Act leaves it to member states to design penalties for most infringements. For breaches of the Article 5 prohibitions, however, the consolidated AI Act sets administrative fines of up to EUR 35 million or, for an undertaking, up to 7 percent of worldwide annual turnover, whichever is higher. That is the top tier in the Act.

The omnibus also changed who supervises some of the biggest systems. Under amended Article 75, the AI Office, the Commission’s own body, is exclusively competent for AI systems built on a general-purpose model by the same provider, and for AI systems that are or are integrated into a very large online platform or search engine designated under the Digital Services Act, with listed exceptions. I read that as meaning that a nudifier function inside a designated platform would be an AI Office matter, not a national one, but I have not seen guidance on this and treat it as an open question.

There is also a timing issue. The Act’s transparency rules for AI-generated content, which require synthetic images to be marked in a machine-readable way, apply to systems already on the market from December 2, 2026 as well. The two obligations land on the same day.

How this fits with the rest of the map

The AI Act ban is one layer in a crowded picture. On the UK side, our summary of police guidance on nudify tools covers the offence side, and the Grok case under Ofcom’s eye shows regulators dealing with one large system. Platform gatekeepers have acted too: Apple’s rules on nudify bait-and-switch apps are an example of private enforcement. In Europe, Germany and France are moving on sexual deepfakes in criminal law.

What the AI Act adds is a supply-side rule. It does not wait for a victim to find the image. It says that a tool whose purpose, or whose foreseeable and unguarded behavior, is to strip or sexualize real people should not be sold in the single market at all. For victims of non-consensual intimate imagery and the wider phenomenon of deepfake porn, that shifts some of the burden from the person harmed to the company that built the tool.

What to watch before and after December 2

Three questions will decide whether the ban bites.

First, what counts as “adequate.” Providers of general-purpose image tools will argue that their guardrails meet the state of the art; regulators will have to decide whether a determined user’s workaround counts as “foreseeable circumvention.” The recitals point to a high bar, but that is guidance rather than a test.

Second, how identifiability is applied. If a system makes sexual images of a real person from a text prompt with a name, the case is clear. If it makes a face that looks a lot like a celebrity, it is not.

Third, who actually enforces. Market surveillance authorities differ widely in size and experience across the member states, and the AI Office is new. A prohibition that nobody has the capacity to enforce is a statement of values rather than a market rule. The text gives regulators strong fines. Whether it gives them the people to use them is not something I could verify.

For adult platforms the practical lesson is narrower than the headlines imply. Consensual, fictional and clearly non-identifiable content is outside the ban. What is inside it is any feature that lets users make explicit images of real people who have not agreed, and any system where that outcome is easy, repeatable and lightly guarded. Operators of such features will want to read Article 5, point (ba), again very carefully before December 2.

Sources

European Parliament press release, “AI Act: EP approves simplification measures and nudifier app ban,” June 16, 2026.
Council of the EU press release, “Artificial Intelligence: Council gives final green light to simplify and streamline rules,” June 29, 2026.
Regulation (EU) 2026/1744 of 8 July 2026 (Digital Omnibus on AI), Article 1 points 7 and 40, and recitals 10 to 16.
AI Act consolidated text as shown by the AI Act Service Desk: Article 5, Article 99(3) and Article 113.