Prohibited AI Practices Under Article 5
The ten AI practices Article 5 bans, who must comply, the narrow exceptions, and fines up to EUR 35 million or 7% of worldwide turnover, whichever is higher.
Article 5 bans ten AI practices: placing certain AI systems on the market, putting them into service or using them. Operators include providers and deployers.
Fines reach EUR 35 million or 7% of total worldwide annual turnover, whichever is higher (Article 99(3)). Systems used exclusively for military, defence or national security purposes are outside the Act (Article 2(3)). Next step: check each of your AI systems against the ten practices below.
Eight bans apply from 2 February 2025; two more from 2 December 2026.
Fines apply from 2 August 2025 and 2 December 2026, respectively.
Ten AI practices are banned
Subliminal and manipulative techniques
Art. 5(1)(a)AI systems that use subliminal techniques beyond a person’s consciousness, or purposefully manipulative or deceptive techniques, to materially distort behaviour. The ban applies when this causes, or is reasonably likely to cause, significant harm.
Exploiting vulnerabilities
Art. 5(1)(b)AI systems that exploit vulnerabilities due to age, disability or a specific social or economic situation, to materially distort behaviour in a way that causes, or is reasonably likely to cause, significant harm.
Non-consensual intimate material
Art. 5(1)(ba)AI systems that generate or manipulate realistic images, videos, audio or similar material of an identifiable person’s intimate parts, or of that person in sexually explicit activities, without their freely given, specific, informed, unambiguous and explicit consent. Manipulation that does not increase the exposure of intimate parts or alter the nature of the activities does not count (Article 5(1b)). Providers are covered only if this is the system’s intended purpose, or if its design, training, architecture, capabilities or user-facing functions make it a reasonably foreseeable and reproducible outcome without significant technical modification and the system lacks reasonable and adequate safeguards to prevent and correct it. Deployers are covered when they use a system for this purpose (Article 5(1a)). Applies from 2 December 2026.
Child sexual abuse material
Art. 5(1)(bb)AI systems that generate or manipulate material or performances within the meaning of Article 2, points (c) and (e), of Directive 2011/93/EU, unless a ‘without right’ defence applies under national law. The same conditions for providers and deployers apply as for point (ba) (Article 5(1a)). Applies from 2 December 2026.
Social scoring
Art. 5(1)(c)AI systems that evaluate or classify people over time based on their social behaviour or personal characteristics, when the score leads to detrimental treatment in unrelated social contexts, or to treatment that is unjustified or disproportionate. The ban is not limited to public authorities.
Predicting crime from profiling
Art. 5(1)(d)AI systems that assess or predict whether a person will commit a criminal offence based solely on profiling or on personality traits. AI that supports a human assessment based on objective, verifiable facts directly linked to a criminal activity is not banned.
Untargeted scraping of facial images
Art. 5(1)(e)AI systems that create or expand facial recognition databases through untargeted scraping of facial images from the internet or CCTV footage.
Emotion recognition at work and in education
Art. 5(1)(f)AI systems that infer the emotions of a person in the workplace or in educational institutions. The ban does not apply to systems intended for medical or safety reasons.
Biometric categorisation by sensitive traits
Art. 5(1)(g)AI systems that categorise people individually based on their biometric data to deduce or infer race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation. Labelling or filtering lawfully acquired biometric datasets, and categorising biometric data in law enforcement, are not covered.
Real-time remote biometric identification
Art. 5(1)(h)Law enforcement use of ‘real-time’ remote biometric identification systems in publicly accessible spaces is banned. Three narrow exceptions are covered further down.
Banned practices carry the highest fine tier
Article 99(3) sets this maximum for breaking an Article 5 ban. The 7% arm applies when the offender is an undertaking and refers to the preceding financial year. For SMEs, including start-ups, the fine is up to the lower of the two amounts (Article 99(6)). The fine in each case depends on factors such as the gravity and duration of the infringement (Article 99(7)). Each Member State decides to what extent fines may be imposed on public authorities (Article 99(8)).
Real-time biometric identification has three narrow exceptions
The ban in Article 5(1)(h) does not apply where use is strictly necessary for one of three objectives. Member States may decide whether to allow such use at all, and under which conditions (Article 5(5)).
Three objectives are allowed
Use is allowed only where strictly necessary for: the targeted search for specific victims of abduction, trafficking or sexual exploitation, and for missing persons; preventing a specific, substantial and imminent threat to life or physical safety, or a genuine and present or foreseeable threat of a terrorist attack; or locating or identifying a suspect of an offence listed in Annex II, for a criminal investigation, prosecution or penalty. The offence must be punishable by a custodial sentence or detention order with a maximum of at least four years (Article 5(1)(h)).
Prior authorisation and safeguards are required
Each use needs prior authorisation from a judicial authority or an independent administrative authority whose decision is binding. In a duly justified urgency, use may start without it if authorisation is requested without undue delay and within 24 hours. If it is rejected, use stops and all data and outputs are deleted (Article 5(3)). Authorising a use also requires a completed fundamental rights impact assessment and registration in the EU database. In urgency, registration may follow without undue delay (Article 5(2)). Each use must also be notified to the market surveillance authority and the national data protection authority (Article 5(4)).
Identification after the event is high-risk, not banned
Remote biometric identification that does not run in real time is not banned. It is a high-risk use under Annex III, point 1(a). In a targeted search for a person suspected or convicted of a criminal offence, the deployer must request authorisation from a judicial authority or an administrative authority whose decision is binding and subject to judicial review, before use, or without undue delay and within 48 hours at the latest. This does not apply to the initial identification of a potential suspect based on objective and verifiable facts directly linked to the offence (Article 26(10)).
Frequently asked questions
Does the social scoring ban apply to private companies?
Yes. Article 5(1)(c) bans social scoring systems without limiting the ban to public authorities. It covers operators that place such a system on the market, put it into service or use it, within the scope of Article 2. Article 2 excludes, among others, military, defence and national security uses (paragraph 3), certain third-country public authorities (paragraph 4), scientific research systems (paragraph 6), research and testing before placing on the market (paragraph 8) and purely personal non-professional use by natural persons (paragraph 10).
How do I check my AI systems against the prohibited practices?
Compare each system’s intended purpose and actual use with the ten practices in Article 5(1), and record the result per system. Scanara supports this with a guided Article 5 screening assessment and rule-based code checks for detectable patterns. Intent and context need your judgement, and the decision stays with you.
Scanara prepares evidence from your repositories
Scanara reads your connected GitHub repositories and runs rule-based scans for standard AI and machine-learning patterns. It can miss non-standard code. Separately, it generates draft documentation from templates and your assessment answers as evidence for your compliance work. You make the classification decisions.