Prohibited AI practices under Article 5: the 8 practices banned outright

The eight prohibited practices

• Subliminal, manipulative, or deceptive techniques (Art. 5(1)(a)) that materially distort a person’s behaviour in a way that causes or is reasonably likely to cause significant harm.
• Exploiting vulnerabilities (Art. 5(1)(b)) due to age, disability, or a specific social or economic situation, to materially distort behaviour and cause harm.
• Social scoring (Art. 5(1)(c)), evaluating or classifying people based on social behaviour or predicted characteristics, leading to detrimental or unjustified treatment, including in contexts unrelated to where the data was originally generated.
• Predicting an individual’s risk of committing a crime (Art. 5(1)(d)) based solely on profiling or assessing personality traits, without objective, verifiable facts directly linked to a criminal activity.
• Untargeted scraping of facial images (Art. 5(1)(e)) from the internet or CCTV footage to build or expand facial recognition databases.
• Emotion recognition in the workplace or education institutions (Art. 5(1)(f)), except where used strictly for medical or safety reasons.
• Biometric categorisation inferring protected attributes (Art. 5(1)(g)) (race, political opinions, trade union membership, religious or philosophical beliefs, sex life, or sexual orientation) from biometric data, with a narrow law-enforcement exception for lawfully acquired biometric datasets.
• Real-time remote biometric identification in publicly accessible spaces for law enforcement (Art. 5(1)(h)), except for narrowly defined purposes such as searching for a specific missing person, preventing an imminent threat to life, or identifying a suspect in a serious crime.

Why 'materially distort behaviour and cause harm' matters

Practices (a) and (b) are not blanket bans on persuasive or targeted design, they require the specific combination of material behavioural distortion and resulting (or reasonably likely) significant harm. A recommendation engine or personalised advertising system is not automatically caught; whether it crosses the line is a fact-specific assessment that typically warrants legal review rather than a self-certified conclusion.

The law-enforcement exceptions are narrow, not general

Both the real-time biometric identification ban (h) and the biometric categorisation ban (g) carry law-enforcement exceptions, but they are deliberately narrow: they require prior authorisation, apply only to the specific purposes listed in the Act, and do not extend to general public-safety or private-sector use. Assuming a “security” justification qualifies without checking the exact wording of Art. 5(2)-(7) is one of the most common ways organisations misjudge their exposure here.

Prohibited status overrides any high-risk classification

If a system matches both a prohibited practice and a high-risk criterion under Annex III, the prohibition controls, there is no partial compliance path through the high-risk obligations (Articles 9-15) that rescues an otherwise-banned use case. This is why prohibited-practice screening should always happen before, and independently of, high-risk classification.