tiltott AI-gyakorlat

in English: prohibited AI practice

basicwidespread

A use of artificial intelligence that the EU considers so dangerous to human dignity and freedom that it has banned it outright.

In plain terms

The EU Artificial Intelligence Regulation (AI Act) classifies AI systems by risk. Most can be used freely, riskier ones under strict conditions, and some uses are prohibited altogether. The latter are the banned AI practices.

It is like the traffic code: most rules tell you how to travel safely, but there are a few things you cannot do under any circumstances, such as driving against traffic on a motorway. Likewise, there is no permit you could obtain for prohibited practices.

The ban does not apply to a specific technology, but to how it is used. Facial recognition can unlock your phone, but it is forbidden to build facial recognition databases by indiscriminately scraping face images from the internet or from surveillance camera footage.

The list includes items such as hidden manipulation of people, exploitation of vulnerable situations, scoring people in ways leading to discrimination based on their behavior, or monitoring emotions in the workplace and at school.

How it works

Article 5 of the regulation originally lists eight prohibited practices: harmful manipulation, exploitation of vulnerabilities, social scoring, law enforcement predictive policing based solely on profiling, indiscriminate collection of facial images, workplace and school emotion recognition, biometric categorization inferring sensitive characteristics, and real-time remote biometric identification (typically facial recognition) for law enforcement purposes in public spaces.
The ban also applies to anyone who markets or deploys such systems, and to anyone who uses them. It does not matter whether the company operates within or outside the EU if the system is used here.
Several prohibitions include narrow exceptions. Emotion recognition is permitted, for example, for medical or security reasons. Real-time facial recognition in public spaces is allowed only if the law of the member state permits it and a court or independent authority grants prior authorization, such as for searching for missing persons, preventing terrorist threats, or locating suspects of serious crimes.
Compliance is monitored by national authorities. Fines can reach 35 million euros or 7 percent of the company's annual global turnover, whichever is higher.
The European Commission examines annually whether the list needs to be modified. However, the list itself can only be changed by amending the regulation, through a decision by the European Parliament and the Council. In 2026, EU lawmakers agreed on such an amendment (in the so-called digital omnibus package): a new ban was added to the list for AI systems that generate sexual or intimate content of real persons without consent, or material depicting the sexual abuse of children.

An everyday example

Your workplace wants to introduce software that measures your enthusiasm or stress levels during video calls based on your face and voice. This practice is banned in the EU; your employer cannot use it even if you consent.

Who is affected

everyoneThe bans protect you from being manipulated, scored, or identified based on your face without your knowledge.

employeeYour employer cannot monitor your emotions with AI unless it is for medical or security reasons. → Work & HR

entrepreneur, executiveIf your company uses an AI system for prohibited purposes, the fine can reach up to 7 percent of your revenue, even if the software was purchased from another party.

developerYou must check at the design stage whether your system or AI agent falls under any prohibition, because there is nothing to get approved afterward. → Code

A common misconception

MythThe EU has banned facial recognition and emotion-detecting AI.

In realityIt is not the technology that is banned, but certain uses in specific situations. Your phone's face unlock or a system monitoring driver fatigue remain permitted.

Limits and risks

The wording of the prohibitions is broad in many places (for example, what counts as 'manipulation causing significant harm'), so borderline cases will ultimately be decided by the authorities and the courts.
The regulation does not apply to systems used exclusively for military, defense, and national security purposes, so these prohibitions do not protect there.
Due to exceptions for public space facial recognition, the ban is not complete, and member states may use the opportunity differently.
The list is valid only in the EU: in other countries, the same practices may be legal, and you may encounter them in services there.

Similar, but not the same

High-risk AI system – High-risk systems can be used under strict conditions, while prohibited practices are not allowed under any conditions.
data protection violation (GDPR) – The GDPR regulates the processing of personal data and often permits it with consent, whereas prohibited AI practices do not become lawful even with consent.
unethical AI use – Many things may be ethically questionable, but only practices on the regulation's closed list count as prohibited.

Related concepts

Broader

EU AI Act, risk-based regulation, AI regulation

Narrower

social scoring, emotion recognition, real-time remote biometric identification, predictive policing, subliminal manipulation

On the edge

High-risk AI system, biometric identification, GDPR, fundamental rights impact assessment, AI proficiency

A brief history

The list of prohibited practices was already included in the European Commission's 2021 draft regulation, and the European Parliament expanded it during negotiations. The AI Act entered into force on 1 August 2024, with the prohibitions to be applied as the first element of the regulation from 2 February 2025. In 2026, EU lawmakers agreed on an amendment to the regulation (digital omnibus), which expands the list with a ban on AI systems generating sexual content without consent.

Where things stand

The bans have been mandatory across the entire EU since February 2025, and the penalty rules have applied since August 2025. The Commission's February 2025 guidelines help with interpretation, but established regulatory and court practice is still developing. The 2026 amendment adds a new ban to the list (producing sexual content without consent).

Status: 2026 November

Also known as

tiltott MI-gyakorlat · prohibited artificial intelligence practice · tiltott AI-rendszer · unacceptable-risk AI · prohibited AI practices · banned AI practices · unacceptable risk AI · AI Act 5. cikk

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.