AI Ethics and Policy 2026-11-19

EU bans AI that strips people in images from 2 December

Short film from Teknikvärlden (in Swedish). The voice is synthetic.

In less than two weeks, on 2 December 2026, it becomes illegal in the EU to sell, deploy or use AI systems that generate or manipulate realistic images, video or audio of an identifiable person's intimate parts, or of that person engaged in sexual activity, without their consent. The same day, AI that produces child sexual abuse material is banned too. The European Commission names the kind of tool it has in mind: so-called nudification apps.

Both bans are new additions to the AI Act. They came in through the AI Omnibus, an amending package that the European Parliament and the Council agreed on politically on 7 May and that entered into force on 27 July 2026. The Act already banned eight AI practices, including social scoring and emotion recognition in workplaces and schools, and those have applied since February 2025. This makes nine.

What the text actually says

Article 5 is more precise than the headlines. Consent has to be freely given, specific, informed, unambiguous and explicit. For providers, the ban applies if the tool is designed to produce this kind of material. It also applies if such output is a reasonably foreseeable and reproducible result of how the system is built and trained, and there are no adequate technical safeguards that reliably prevent it. In other words, it is not enough for a provider to point to how the app is marketed.

For users, the ban kicks in when they use a system to generate the material themselves. An edit that doesn't expose more of someone's intimate parts, or change the nature of the sexual activity shown, does not count as manipulation under the Act.

Breaching the prohibitions carries the Act's heaviest fines: up to €35 million or 7% of total worldwide annual turnover, whichever is higher. Most AI systems are policed by authorities that each member state designates.

2 December is also the deadline for providers of existing systems to meet the marking and detection obligation in Article 50(2).

High-risk rules pushed back

The same package delayed other parts of the Act. Rules for high-risk AI in biometrics, critical infrastructure, education, employment, migration, asylum and border control now apply from 2 December 2027. AI built into products such as lifts and toys has until 2 August 2028. The Commission says the staggered timeline is meant to give technical standards and other support tools time to be ready before the rules bite.

Sources

There is a short film on this topic on the Teknikvärlden YouTube channel.