The EU Bans Nudify Apps From December 2. What the Rule Covers
The EU AI Act's ninth banned practice starts in December 2026. What it forbids, who is liable, and how it fits the watermarking and high-risk deadlines.

Most of the AI Act’s banned practices have been in force since February 2025. One, added this year, has not started yet. On December 2, 2026, the European Union begins to prohibit AI systems built to produce non-consensual sexual imagery, the category usually called “nudification” or “nudify” apps. The Commission’s own page counts it as the ninth of the Act’s prohibited practices, and the Commission says it was introduced as part of the AI Omnibus, an amendment passed this year.
Two weeks out, the question is not whether the rule exists but what it actually forbids, who is on the hook, and how it sits next to the other AI Act deadlines that moved this year. Here is how the pieces fit.
Where this rule came from
The AI Act itself, Regulation (EU) 2024/1689, entered into force on 1 August 2024 and became applicable on 2 August 2026, with exceptions. The nudification ban is not in the original text. It arrived through the “AI Omnibus,” a package the Commission proposed on 19 November 2025 to simplify how the Act is implemented.
According to the Commission, a political agreement between Parliament and Council came on 7 May 2026 and the Omnibus entered into force on 27 July 2026. Most of the package is about easing compliance: extended timelines, simplified obligations for small mid-cap companies, and more access to regulatory sandboxes. The ban is the part that adds a hard limit rather than loosening one. In the European Parliament’s press release, one of its lead negotiators describes the ban as a key part of Parliament’s mandate.
What the ban prohibits
The Parliament’s summary of the deal describes the scope. The prohibition covers AI systems that create child sexual abuse material, or that depict the intimate parts of an identifiable person, or show them engaged in sexually explicit activities, without that person’s consent. The content can be images, video or audio.
It attaches to three different actions:
- placing an AI system on the EU market with the purpose of creating such content;
- placing a system on the EU market without reasonable safety measures to prevent such creation;
- deployers using these systems for the purpose of creating such content.
The second item is the broader one: by its wording, a system does not have to be built for this purpose to be caught if it reaches the market without reasonable safeguards against it. The Parliament release does not say what counts as “reasonable” in practice, so that is left to guidance and enforcement. The third item reaches the people who operate these systems, not only the companies that sell them.
The Commission’s page frames the same rule in plain terms: AI systems that generate non-consensual sexually explicit and intimate content or child sexual abuse material, “such as AI ’nudification’ apps.” Both institutions give December 2026 as the start. The Parliament says companies have until 2 December 2026 to bring their systems in line.
Why December and not August
The AI Act as a whole became applicable on 2 August 2026, so a December start for this rule looks odd. The explanation is that the ban was folded in as an amendment, with a transition period built in. The unofficial AI Act timeline site artificialintelligenceact.eu lists the new prohibitions, in Article 5, as becoming applicable on 2 December 2026. That site is a reference tool rather than an official publication, so treat the exact article numbering as a pointer and check the Official Journal text before citing it.
The same date for watermarking
December 2 is not only about nudification. The Parliament release says the deal also delays the application of watermarking obligations for AI-generated content until 2 December 2026, instead of the 2 February 2027 date in the Commission’s original proposal. Watermarking techniques, in its words, allow for the detection and tracing of AI-generated content.
There is a nuance here. The Commission’s overview says the AI Act’s transparency rules take effect in August 2026: chatbots must tell people they are talking to a machine, and providers of generative AI must make AI-generated content identifiable, with deepfakes and certain public-interest text clearly labelled. The timeline site clarifies the December date: providers of systems already on the market before 2 August 2026 have until 2 December 2026 to comply with the marking requirement in Article 50(2).
That is a good place to connect the rule with the practical problem it addresses. Marking and labelling is one defence, but detectors and provenance data have limits of their own, which we cover in our guide to verifying synthetic media.
What did not move
The Omnibus is mostly known for delays, so it is worth being clear about what it postponed and what it did not. According to the Commission, rules for high-risk systems in areas such as biometrics, critical infrastructure, education, employment, migration, asylum and border control now apply from 2 December 2027. High-risk AI built into regulated physical products, such as machinery, toys and lifts, follows on 2 August 2028. The Commission says the sequencing is meant to make sure standards and other support tools are in place first.
The earlier prohibitions were not touched. Prohibitions one to eight, including social scoring and untargeted scraping of faces to build recognition databases, have applied since February 2025, and the nudification ban is added alongside them rather than replacing anything.
Who enforces it
Since 2 August 2026, the AI Office and the authorities of the member states are responsible for supervising and enforcing the AI Act. The Commission says the AI Office holds enforcement powers over general-purpose AI models, and the Omnibus extends its oversight to certain systems built on those models and embedded in large online platforms and search engines.
What to take from it
Three points are worth keeping. First, the ban targets the purpose and safeguards of a system, and also the person who uses it, not just a category of product. Second, the December 2 date is shared with a watermarking deadline, so the next few weeks are a compliance moment for any company whose tools generate images, video or audio. Third, the headline delays in the Omnibus, to 2027 and 2028, apply to high-risk systems and do not soften this particular rule.
What the sources do not say is how the ban will be tested in practice, which safeguards will be judged “reasonable,” or what penalties will apply to violations. Those answers will come from the Commission’s guidance and from the first enforcement cases, and the primary documents linked above are the place to watch for them.