More info

AI Bans in the EU: What Has Been Prohibited Since February 2025

Since 2 February 2025, certain AI practices have been banned across the EU. The German Federal Network Agency lists the eight categories in Article 5 of the AI Act.
Zwei Studentinnen auf dem Campus im Herbst

Contents

Since early 2025, the AI Regulation of the European Union has become part of everyday life for universities, employers and providers. One part applies immediately: certain practices involving artificial intelligence have been prohibited since 2 February 2025. This guide explains which practices those are, based on information from the German Federal Network Agency (Bundesnetzagentur).

Short answer: The Federal Network Agency states: “These prohibitions have applied since 2 February 2025.” Article 5(1) of the AI Regulation lists eight categories, labelled a to h, of prohibited practices. Real-time remote biometric identification in publicly accessible spaces for law enforcement purposes is permitted only under narrow conditions. Last updated: November 2026.

The most important points at a glance

  • According to the Federal Network Agency, the prohibitions have applied since 2 February 2025.
  • Article 5(1) of the AI Regulation names eight prohibited practices, marked with the letters a to h.
  • Real-time remote biometric identification in publicly accessible spaces for law enforcement is allowed only with a national authorisation that meets further requirements.
  • The prohibitions concern the use of certain systems, not every AI application with similar features.
  • Whether a specific product falls under them depends on its function and its use.

Since when do the AI bans apply?

The Federal Network Agency gives the cut-off date clearly: the prohibitions apply since 2 February 2025. From that point on, providers and deployers had to observe the requirements listed in this article. Later obligations, such as labelling duties, must be viewed separately.

For you as a user, the cut-off date is mainly a benchmark. An application that implements a prohibited practice should not have been offered or used since that date. Whether a particular service complies can often not be checked fully from the outside.

Which eight practices are prohibited under Article 5?

The letters show which categories are covered. The table gives English translations of the designations that the Federal Network Agency uses.

LetterProhibited practice (as named by the Federal Network Agency)
aHarmful manipulation and deception
bExploiting vulnerabilities due to age, disability or social or economic situation
cSocial scoring
dIndividual risk assessment and prediction of criminal offences based solely on profiling
eUntargeted scraping of facial images to build facial recognition databases
fInferring emotions in the workplace and in educational institutions
gBiometric categorisation by sensitive characteristics
hReal-time remote biometric identification in publicly accessible spaces for law enforcement

The letters do not name products. They describe behaviours or purposes, which is why the concrete use matters.

What does this mean for students in everyday life?

The prohibitions address providers and deployers, not students who use a tool privately. Still, students come across systems that may fall into these areas. Letter f is relevant, for example, when universities or employers want to evaluate emotions from faces or voices. In that case, it is worth asking what function a system actually performs.

Conversely, a mention of “AI” in advertising does not make an offer prohibited, and a ban under Article 5 does not mean that every analysis using AI is unlawful. What counts is which practice is carried out and by whom.

The exception for real-time remote identification

Letter h concerns real-time remote biometric identification in publicly accessible spaces for law enforcement purposes. According to the Federal Network Agency, its use without a national authorisation that meets the requirements of paragraphs 2 to 7 has been prohibited since 2 February 2025. The exception is therefore narrow and bound to conditions that not every authority can meet.

In practice, such systems should be used only where the legal framework expressly permits it. Under the wording, a single measure without authorisation is prohibited.

How to recognise prohibited practices in daily use

Even without specialist knowledge, some patterns can point to one of the prohibited practices. The list below translates the letters into observations you can make when working with software. It does not replace a legal assessment.

  • The application assesses people based on their behaviour across different areas of life (letter c).
  • The application estimates the emotions of employees or learners, for example from facial expressions or voice (letter f).
  • The application matches faces untargeted against images from the internet to build a database (letter e).
  • The application deliberately exploits a person’s characteristics because they are young, disabled or in a difficult economic situation (letter b).

These signs are prompts for reflection. Whether a specific system implements a prohibited practice depends on its precise function, purpose and context. With systems that look harmless in advertising, asking about the technical purpose is especially worthwhile.

Universities and the workplace

Letter f explicitly names educational institutions. For students, this can become relevant when learning platforms or examination systems evaluate mood or concentration from video or audio data. Ask the university whether such evaluations take place and what function the system has before you use it.

The same applies to employers. If you work in a part-time job or an internship with software that estimates the emotions of staff, find out which data is processed. The prohibitions target exactly this area, which is why the question of purpose matters so much.

Data protection and AI bans are separate questions

A tool can be unremarkable under data protection law and still implement a prohibited practice, and conversely an application can raise data protection questions without falling under any prohibition. Both checks should therefore be carried out separately. For data protection, topics such as consent and access play a role, while for the AI bans the question is the practice itself.

The AI Regulation also sets obligations that do not begin with the prohibitions. The prohibitions are the early, clearly dated part. Anyone relying on a service should therefore also watch labelling duties, which the guide AI images: which labels exist and what they are worth (in German) explains.

Frequently asked questions

Does the prohibition also apply to private individuals?

The prohibitions in Article 5 are directed, according to the structure of the regulation, at providers and deployers of AI systems. A private person who uses a finished tool for their own purposes is typically not the focus of these duties. The precise responsibilities follow from the wording of the regulation.

Is an AI application prohibited if it recognises emotions?

Not every emotion recognition is prohibited. Letter f concerns inferring emotions in the workplace and in educational institutions. Outside these contexts, other rules apply. Whether a specific system falls under the ban depends on its function and place of use.

Which prohibitions apply now, and which later?

The Federal Network Agency names 2 February 2025 as the start date for the prohibitions. Which further obligations of the regulation apply on which date is answered by the regulation itself, and those dates are not the subject of this guide.

What happens if I use a prohibited application?

The prohibitions address providers and deployers. For users, this usually does not create a direct ban, but it does mean you can avoid offers that obviously violate the requirements. Legal consequences in an individual case depend on the facts.

Where can I find the wording of Article 5?

The text of the AI Regulation is published in the Official Journal of the European Union. The Federal Network Agency summarises the prohibitions clearly and refers to the text of the regulation.

What to do now

  1. Before you use an AI offer for study or work, check which function it actually performs.
  2. Ask your university or employer whether a tool evaluates emotions or biometric features.
  3. Note the provider’s information on data protection and purpose.
  4. If you suspect a violation, contact the competent body, for example the Federal Network Agency.

Related reading

Sources: Federal Network Agency, prohibited practices under the AI Regulation (German, accessed 2 November 2026)

Last updated: November 2026. This guide is general information, not individual advice.

Teilen

Weitere Artikel

Student vergleicht am Schreibtisch zwei Versichertenkarten

Gatekeeper model over 30: what it means for costs and fund choice

Student vergleicht am Schreibtisch zwei Versichertenkarten

Requesting a contribution certificate from your health insurer

Student vergleicht am Schreibtisch zwei Versichertenkarten

Fund merger or closure: what happens to contribution and membership