Prohibited AI practices under the AI Act: eight things that may not be done at all

In brief

The AI Act does not regulate all AI the same way. Most use is free, some is high-risk use with obligations, and a small set of practices is prohibited outright. The prohibitions are in Article 5, there are eight of them, and they have applied since 2 February 2025, the first of the whole regulation. They concern manipulation, exploitation of vulnerable groups, social scoring, predicting criminal risk by profiling, scraping facial images, emotion recognition in workplaces and education, biometric categorisation and real-time remote biometric identification. For an ordinary organisation the list is a short check, but it must be made: a prohibition cannot be bought away as a finished product.

The eight prohibitions

Article 5(1) prohibits placing on the market, putting into service and using the following AI systems.

Point	Prohibited practice	What it means in practice
a	Subliminal, purposefully manipulative or deceptive techniques that materially distort behaviour by impairing the ability to make an informed decision and cause significant harm	An interface or content that steers a person to a decision they would not otherwise make, in a way they do not notice
b	Exploiting vulnerabilities of age, disability or a specific social or economic situation to distort behaviour, causing significant harm	A system aimed at children, older people or the indebted that uses their situation against them
c	Social scoring based on social behaviour or personal characteristics, where it leads to detrimental treatment out of context or disproportionately	Scoring where behaviour in one service worsens treatment in another
d	Assessing or predicting the risk of committing a criminal offence based solely on profiling or personality traits	Does not cover supporting a human assessment based on objective and verifiable facts
e	Creating or expanding facial recognition databases through untargeted scraping of facial images from the internet or CCTV	A service built from collected facial images without targeting
f	Inferring emotions in the workplace and in educational institutions	Exception only for medical and safety purposes
g	Biometric categorisation to deduce race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation	Does not cover labelling lawfully acquired datasets in law enforcement
h	Real-time remote biometric identification in publicly accessible spaces for law enforcement	Allowed only where strictly necessary, in listed cases, with prior authorisation

What prohibitions a to c share is that they require significant harm or unjustified treatment; they do not prohibit persuasion or personalisation as such. Prohibitions e to h concern biometrics, and of these f and g are the ones an ordinary employer may meet in procurement.

Emotion recognition in the workplace

The prohibition most likely to touch an ordinary organisation is point (f): using an AI system to infer a natural person's emotions in the workplace and in educational institutions is prohibited, except where the use is intended for medical or safety reasons.

Products on the market do exactly this under another name: emotion analysis of customer service calls that also measures the employee's voice; an attention or engagement meter for video meetings; facial expression analysis in recruitment interviews; attention tracking on a learning platform. If the system infers an employee's or a pupil's emotional state, it falls within the prohibition regardless of whether it is called wellbeing or feedback. The line to the safety exception is narrow: detecting a driver's fatigue is safety, detecting an employee's dissatisfaction is not.

Where a system may be used, for instance aimed at a customer outside the workplace, Article 50(3) requires informing those exposed to it of its operation. That is covered in Article 50 of the AI Act.

Prohibited, high-risk, other

Prohibition is the top step of the risk classification. Below it are high-risk systems, which may be used subject to obligations, and below them everything else. The same purpose can be on a different step depending on a detail: screening job applications is high-risk use under Annex III, but if the same system infers an applicant's emotions from a video interview, it is prohibited. Assessing criminal risk is prohibited when based solely on profiling, and high-risk use when it supports a human assessment based on verifiable facts.

That is why our skills catalogue treats risk classification and prohibitions as one skill. At level 1 everyone knows that some uses are prohibited. At level 3 the AI officer makes a risk classification assessment and a prohibition check for the organisation's own systems, and at level 4 resolves the edge cases and roles. The high-risk obligations are covered in High-risk AI.

Dates and penalties

Under Article 113, Chapters I and II, to which Article 5 belongs, have applied since 2 February 2025. Regulation (EU) 2026/1744 postponed the high-risk obligations but not the prohibitions; it added new points to Article 5(1) and new paragraphs to the article that apply from 2 December 2026. Their content must be read from the Omnibus regulation itself, and the AI officer's prohibition check is updated when they become applicable.

The penalties are the regulation's harshest. Under Article 99, non-compliance with the prohibitions of Article 5 can be fined up to 35 million euros or 7 percent of the total worldwide annual turnover of the preceding financial year, whichever is higher; for SMEs and start-ups, under Article 99(6), the lower of the two. For the public sector, Member States lay down the penalties applicable to public authorities.

What the prohibitions do not cover

The prohibitions do not cover personalisation, recommendation or marketing as such, only techniques that materially distort behaviour and cause significant harm. They do not cover measuring customer satisfaction with a survey, only inferring emotions with AI in the workplace. They do not ban facial recognition in every situation, only untargeted scraping and real-time remote identification in public spaces for law enforcement. And Article 5(8) recalls that the article does not affect prohibitions that come from other Union law: the GDPR and the prohibitions of discrimination remain in force alongside it.

The prohibition check is the AI officer's first task and the first line of the use policy. The AI officer's role is described in Who in the organisation is responsible for AI use? and the use policy in An AI use policy for the workplace.