The EU AI Act's Prohibited Practices (Article 5): The Eight Banned Uses
The eight prohibited AI practices under Article 5 of the EU AI Act, in force since 2 Feb 2025. Fines up to €35M or 7%, plus a table and example.
Some AI uses are not regulated — they are banned outright. Article 5 of the EU AI Act, Regulation (EU) 2024/1689, lists eight prohibited practices that no one may place on the market, put into service, or use in the EU. These have been in force since 2 February 2025, and they carry the regulation's heaviest fine: up to €35 million or 7% of worldwide annual turnover (Article 99(3)). If a use falls in Article 5, no amount of documentation or oversight makes it lawful — you stop.
The hard part for most companies is not the prohibited uses themselves but the lawful look-alikes that sit just beside them. This guide lists the eight, then draws the lines that actually matter.
The eight prohibited practices
| # | Article 5(1) | Prohibited practice |
|---|---|---|
| 1 | (a) | Subliminal, manipulative or deceptive techniques that materially distort behaviour and cause significant harm |
| 2 | (b) | Exploiting vulnerabilities of age, disability or a specific social/economic situation to distort behaviour and cause significant harm |
| 3 | (c) | Social scoring — evaluating or classifying people over time from social behaviour or personal traits, leading to detrimental treatment |
| 4 | (d) | Predicting criminal offending based solely on profiling or personality traits (not on objective, verifiable facts tied to criminal activity) |
| 5 | (e) | Untargeted scraping of facial images from the internet or CCTV to build or expand facial-recognition databases |
| 6 | (f) | Emotion recognition in the workplace and education settings (save for medical or safety reasons) |
| 7 | (g) | Biometric categorisation that infers sensitive attributes — race, political views, trade-union membership, religion, sex life, sexual orientation |
| 8 | (h) | Real-time remote biometric identification in publicly accessible spaces for law-enforcement purposes (narrow, authorised exceptions only) |
These are bans on the use, not on a tool or a brand. The same underlying technology — face recognition, sentiment analysis, scoring — can be lawful in one deployment and prohibited in another. Always classify by what the system is used for.
The lawful look-alikes (where companies get it wrong)
Credit scoring is not social scoring. Assessing a person's creditworthiness is high-risk (Annex III, point 5(b)), not prohibited. Social scoring under 5(1)(c) is the cross-context evaluation of people's general social behaviour leading to unjustified or disproportionate detrimental treatment — for example, denying an unrelated public benefit because of unrelated private conduct. A bank scoring a loan applicant on relevant financial data is doing regulated high-risk AI, not a banned practice.
One-to-one verification is not real-time remote biometric identification. Unlocking a phone or confirming an employee's identity at a turnstile (1:1 matching, with consent, against a known template) is not the 5(1)(h) prohibition. The ban targets real-time, remote, one-to-many identification of people in public spaces for law enforcement. Most workplace and consumer biometrics are verification, not remote identification.
Emotion recognition outside work and school can be lawful. The 5(1)(f) ban is specific to workplace and education. The same capability used elsewhere may be lawful — though it can be high-risk biometrics (Annex III, point 1) and usually triggers an Article 50(3) notice that the person is being analysed. A common error is to call workplace emotion recognition "high-risk" — it is prohibited.
Timing and the new 2 December 2026 ban
The eight prohibitions have applied since 2 February 2025. They were not affected by the Digital Omnibus deferral of the high-risk regime — they are live now.
A new prohibition arrives on 2 December 2026: AI used to generate or manipulate child sexual abuse material (CSAM), and so-called "nudifier" applications. Treat it as a near-term addition to the Article 5 list rather than something on a distant horizon.
A note on the wider timeline: the high-risk Annex III deadline, provisionally agreed on 6–7 May 2026, has been adopted to move to 2 December 2027 — the European Parliament voted on 16 June 2026 and the Council adopted it on 29 June 2026, with entry into force pending Official Journal publication. None of that touches Article 5, which is already in force.
Worked example
Nordkran GmbH, a 140-person logistics firm in Hamburg (€31M turnover), is weighing three AI ideas. (1) A camera system that reads warehouse staff's facial expressions to flag "disengagement" for managers. (2) A 1:1 face-scan to open secure doors for badged employees. (3) A model that scores delivery-route risk from traffic and weather data.
Verdicts: idea (1) is prohibited — emotion recognition in the workplace, Article 5(1)(f); it cannot launch, full stop. Idea (2) is lawful biometric verification, not the 5(1)(h) remote-identification ban — proceed with the usual data-protection care. Idea (3) is not biometric or prohibited at all; route risk from operational data is, on these facts, minimal-risk AI. One product idea dies at the prohibition line; the other two proceed. That is the value of classifying each use before building it.
How Confir helps
Confir screens every registered use against the eight Article 5 categories first, before any high-risk analysis, and flags a prohibited use with the exact sub-paragraph and the reason it fired — so a banned idea is caught at the design stage, not after launch. The engine is deterministic and rule-based: the same intake always returns the same verdict, and you can see which rule applied.
Related guides
- EU AI Act Article 5, explained
- Prohibited AI practice, defined
- Unacceptable-risk AI
- Real-time biometric identification prohibition
- Social scoring AI
Manage your EU AI Act compliance in one place
Confir automates risk classification, technical documentation, and audit trails for any company. No consultants. No 6-month projects. 14-day free trial.
Start free trial →