Prohibited AI practices — EU AI Act Article 5

The eight prohibited categories

1. Art 5(1)(a) — Subliminal, manipulative or deceptive techniques. AI systems that deploy subliminal techniques beyond a person’s consciousness or purposefully manipulative or deceptive techniques, with the objective or the effect of materially distorting behaviour by appreciably impairing the ability to make an informed decision, thereby causing or being reasonably likely to cause significant harm.
2. Art 5(1)(b) — Exploitation of vulnerabilities. AI systems that exploit any of the vulnerabilities of a natural person or a specific group of persons due to their age, disability or a specific social or economic situation, with the objective or the effect of materially distorting their behaviour in a manner that causes or is reasonably likely to cause significant harm.
3. Art 5(1)(c) — Social scoring. AI systems for the evaluation or classification of natural persons or groups over a certain period of time based on their social behaviour or known, inferred or predicted personal or personality characteristics, with the social score leading to detrimental or unfavourable treatment in social contexts unrelated to those in which the data was generated, or unjustified or disproportionate treatment.
4. Art 5(1)(d) — Predictive policing based solely on profiling. AI systems for making risk assessments of natural persons to assess the risk of them committing a criminal offence, based solely on profiling or on assessing personality traits and characteristics. Carve-out for systems used to support human assessment of the involvement of a person in a criminal activity that is already based on objective and verifiable facts.
5. Art 5(1)(e) — Untargeted facial-image scraping. AI systems that create or expand facial-recognition databases through the untargeted scraping of facial images from the internet or CCTV footage.
6. Art 5(1)(f) — Emotion recognition in workplaces and educational institutions. AI systems used to infer emotions of a natural person in the areas of workplace and education institutions. Narrow carve-out for systems intended to be put in place or into the market for medical or safety reasons.
7. Art 5(1)(g) — Biometric categorisation by sensitive attributes. Biometric categorisation systems that categorise individually natural persons based on their biometric data to deduce or infer their race, political opinions, trade-union membership, religious or philosophical beliefs, sex life or sexual orientation. Narrow carve-outs for labelling and filtering of lawfully acquired biometric datasets and for categorisation of biometric data in the area of law enforcement.
8. Art 5(1)(h) — Real-time remote biometric identification in public for law enforcement. The use of ‘real-time’ remote biometric identification systems in publicly accessible spaces for the purposes of law enforcement, with three narrow exceptions subject to the safeguards in Art 5(2)–(7) (targeted victim search, prevention of a specific and present threat including terrorist attack, identification of a suspect of serious crime).

Why the workplace emotion-recognition ban matters for HR-tech

Article 5(1)(f) is the prohibition most relevant to HR-tech deployers. It applies in the workplace regardless of the seniority of the staff, the consent of the data subject, or the commercial intent. It captures, for example, AI features that infer happiness, stress, anger or engagement from voice, facial expression or body language during interviews, performance reviews or routine workplace monitoring.

The medical-or-safety carve-out is narrow: the system must be intended to be put in place for those purposes, not merely capable of being used that way. Wellbeing dashboards and engagement analytics that infer emotional state do not qualify for the carve-out.

Recital 18 and Recital 44 distinguish emotion inference (banned in workplace and education) from detection of physical states such as fatigue (not banned). A drowsiness detector for a truck-driver safety system can therefore continue; an interview tool that scores the candidate’s emotional state cannot.

Commission guidelines — February 2025

On 4 February 2025 the European Commission published Guidelines on prohibited artificial intelligence practices established by Regulation (EU) 2024/1689. The Guidelines elaborate on each Article 5 sub-paragraph, the carve-outs, and the relationship to other Union law (GDPR, Law Enforcement Directive, Charter of Fundamental Rights). They are non-binding but reflect the Commission’s enforcement stance.

Penalty exposure (Art 99(3))

Article 99(3) sets the top tier of fines for Article 5 breaches: up to EUR 35 000 000 or, if the offender is an undertaking, up to 7 % of its total worldwide annual turnover for the preceding financial year, whichever is higher. Article 99(6) sets a lower cap (the lesser of the two amounts or percentages) for SMEs including start-ups.

Member State supervisory authorities apply Article 5 enforcement at national level; for certain practices (Article 5(1)(h)) authorisation regimes operate prior to use.

Deployer screening checklist

• For every AI system in the inventory: is any Article 5 trigger present? Documented yes/no with reasoning.
• HR-tech systems: explicit check for emotion-inference features (Art 5(1)(f)).
• Engagement / wellbeing tools: confirm they do not infer emotional state.
• Vendor contract clause requiring the supplier to confirm no Article 5 practice is in scope.
• Process for stopping use if a prohibited feature is later added by the vendor.
• Article 4 training coverage that lists the eight Article 5 categories.

Common misconceptions

• “Consent makes it OK.” Article 5 prohibits the practice; data-subject consent does not unlock it.
• “Emotion AI is fine if it’s marketed as wellbeing.” If the system infers emotional state of staff in the workplace, the label is irrelevant — Art 5(1)(f) applies.
• “Article 5 only catches providers.” The prohibition covers placing on the market, putting into service and use. Deployers that use a prohibited system are equally exposed.
• “Article 5 starts in August 2026.” No. Article 5 has been enforceable since 2 February 2025 (Article 113(a)).

Related EU guides

• AI literacy obligation — Article 4
• EU AI Act conformity assessment
• Annex IV technical documentation
• NIS2 incident reporting timeline
• GPAI vs deployer obligations

Sources

• Regulation (EU) 2024/1689, Articles 5, 99(3), 99(6), 113(a); Recitals 18 and 44 — EUR-Lex: eur-lex.europa.eu/eli/reg/2024/1689/oj
• European Commission — Guidelines on prohibited AI practices (4 February 2025): digital-strategy.ec.europa.eu
• AI Act Service Desk — Article 5: ai-act-service-desk.ec.europa.eu/en/ai-act/article-5