Prohibited AI practices under the EU AI Act

Article 5 asks for no risk assessment. It prohibits the placing on the market, the putting into service and the use of a short list of practices, so a company that merely uses such a system is caught as squarely as the provider that built it. The list covers harmful manipulation and exploitation, social scoring, individual crime prediction, untargeted scraping of facial images, emotion recognition at work and in education, biometric categorisation by sensitive attributes, and real-time remote biometric identification for law enforcement. All have been prohibited since 2 February 2025.

The Commission's guidelines C(2025) 5052 final read each prohibition condition by condition, and this site renders them chapter by chapter. Two points belong in any internal assessment. The prohibitions require several cumulative conditions, so a practice that resembles a heading is not automatically banned. And Article 5(8) leaves other Union law untouched, so a practice outside Article 5 can still be unlawful under data protection, consumer or employment law.

Article 5(1)(a) requires four cumulative conditions: the system is placed on the market, put into service or used; it deploys subliminal techniques beyond a person's consciousness or purposefully manipulative or deceptive techniques; those have the objective or effect of materially distorting behaviour by appreciably impairing the ability to make an informed decision; and the distorted behaviour causes or is reasonably likely to cause significant harm. The guidelines place lawful persuasion outside the prohibition.

Article 5(1)(b) has the same structure but targets the person rather than the technique: exploiting vulnerabilities due to age, disability or a specific social or economic situation. A persuasive advertisement or an ordinary personalised offer is not caught. The case aimed at is a system that detects financial distress and uses that state to push someone into a harmful decision. Significant harm is judged by severity, context and cumulative effect, not by the name a vendor gives the feature.

Not automatically. Article 5(1)(c) bans evaluating or classifying people over a period of time on the basis of social behaviour or known, inferred or predicted personal characteristics, where the resulting social score leads to detrimental treatment in social contexts unrelated to those in which the data was originally generated or collected, or to treatment unjustified or disproportionate to the behaviour. The guidelines stress that this applies in both public and private contexts and is not tied to a sector.

Ordinary credit scoring is elsewhere. Evaluating the creditworthiness of natural persons is listed in Annex III, point 5(b) as high-risk, so it is regulated rather than banned, and fraud detection is carved out. The line to watch is the reuse of data across unrelated contexts: a score built from social media behaviour and then used to decide access to an unrelated service sits far closer to the prohibition than a score built from payment history and used for credit.

Article 5(1)(d) prohibits AI that assesses or predicts the risk of a natural person committing a criminal offence based solely on profiling or on assessing personality traits. It does not prohibit systems supporting a human assessment of involvement in criminal activity that is already based on objective and verifiable facts directly linked to it. Article 5(1)(e) separately prohibits creating or expanding facial recognition databases through untargeted scraping of facial images from the internet or CCTV footage, with no exception for private or research use.

Article 5(1)(f) prohibits AI that infers emotions of a natural person in the areas of the workplace and education institutions, except where the system is intended for medical or safety reasons. The guidelines read workplace broadly: any physical or virtual space where people carry out tasks assigned by an employer or the organisation they are affiliated with, independent of status as employee, contractor, trainee or volunteer, and extending to candidates during selection and hiring.

Article 5(1)(g) prohibits biometric categorisation that individually categorises people by their biometric data to deduce race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation. Labelling or filtering lawfully acquired biometric datasets is not covered. Where these uses are not prohibited they are still regulated: both appear in Annex III, point 1 as high-risk, and deployers must inform the people exposed to them under Article 50(3).

Regulation (EU) 2026/1744 added two prohibitions to Article 5(1). Point (ba) covers AI systems that generate or manipulate realistic intimate imagery of an identifiable natural person without consent; point (bb) covers AI systems that generate or manipulate child sexual abuse material. Both apply from 2 December 2026 and sit in the same penalty tier as the rest of Article 5: up to 35 million euro or 7 percent of worldwide annual turnover (Article 99(3)).

For most SMEs the practical task is narrow. Real-time remote biometric identification under Article 5(1)(h) is addressed to law enforcement and needs national law, prior judicial or independent authorisation and a fundamental rights impact assessment (Article 5(2) and (3)), so it rarely arises commercially. What does arise is emotion analytics in HR or customer contact tools, scoring features that reuse data across contexts, and generative image features. Ask each vendor in writing which of these its product performs.