Prohibited vs High-Risk AI Under the EU AI Act
The most-confused EU AI Act boundary: Article 5 prohibited (banned, since Feb 2025) vs Annex III high-risk (allowed with controls), with a comparison table.
This is the boundary that trips up most teams. Prohibited practices (Article 5) are banned outright — no controls make them lawful, and the original list has been in force since 2 February 2025, with two further bans applying from 2 December 2026. High-risk systems (Article 6 + Annex III) are allowed, provided you meet the requirements, and apply on a later timeline. The same subject area — biometrics, scoring, policing, emotion recognition — can land on either side depending on a narrow factual line. Get the side wrong and you either deploy something illegal or build a heavy compliance stack for something that is simply forbidden.
Under Regulation (EU) 2024/1689, the deciding factor is always the specific use, not the broad topic. Here is where each line falls.
The four boundaries that are most confused
| Subject | Prohibited (Article 5) | High-risk / allowed (Annex III) |
|---|---|---|
| Scoring people | Social scoring — evaluating or classifying people over time based on social behaviour or personal traits, leading to detrimental treatment in unrelated contexts (Article 5(1)(c)) | Creditworthiness / credit scoring — assessing a person's credit standing (Annex III point 5(b)); high-risk, allowed with controls (excludes fraud detection) |
| Emotion recognition | Inferring emotions in the workplace or in education (Article 5(1)(f)) | Emotion recognition outside those settings — lawful, often as a high-risk biometric use (Annex III point 1) and/or an Article 50(3) notice |
| Policing / risk | Predicting criminal offending solely from profiling a person or their personality traits (Article 5(1)(d)) | Law-enforcement risk-assessment tools that support human officers with broader evidence (Annex III point 6); high-risk, allowed |
| Biometric ID | Real-time remote biometric identification in publicly accessible spaces for law enforcement (Article 5(1)(h), with narrow exceptions) | Post / other remote biometric identification and biometric categorisation (Annex III point 1); high-risk, allowed with controls |
The pattern: prohibition attaches to the most intrusive, rights-destroying form of a use; the high-risk regime governs the controlled, accountable form of a related use.
Why the difference matters in practice
Different verdicts. A prohibited practice cannot be made compliant — you do not write a risk-management file for it, you do not deploy it. A high-risk system is lawful once you meet the requirements (Article 9 risk management onwards) and your role's duties (Article 16 provider / Article 26 deployer).
Different dates. The original Article 5 prohibitions have applied since 2 February 2025 — they are live now. The Digital Omnibus (Regulation (EU) 2026/1744, in force since 27 July 2026) adds two more from 2 December 2026: AI that generates non-consensual intimate images of identifiable people — broader than "nudifier" apps (Article 5(1)(ba)) — and AI that generates child sexual abuse material (Article 5(1)(bb)). Stand-alone Annex III high-risk obligations apply from 2 December 2027, a date the same Omnibus set (originally 2 August 2026).
Different penalties. A breach of an Article 5 prohibition is the top tier — €35,000,000 or 7% of total worldwide annual turnover, whichever is higher (Article 99(3)). A breach of high-risk obligations sits one tier down — €15,000,000 or 3% (Article 99(4)). For SMEs and start-ups, the fine is the lower of the percentage or the fixed amount (Article 99(6)). Small mid-caps get that lower-of cap only for Article 99(4) and (5) fines — not for an Article 5 breach (Article 99(6a)).
| Prohibited (Art 5) | High-risk (Annex III) | |
|---|---|---|
| Status | Banned outright | Allowed with controls |
| In force | 2 February 2025 (Art 5(1)(ba)/(bb): 2 December 2026) | 2 December 2027 |
| Max penalty | €35M or 7% (Art 99(3)) | €15M or 3% (Art 99(4)) |
| Your job | Do not deploy | Meet the requirements + your role's duties |
How to tell the side: ask what the system actually does
Three questions settle most cases. Does it score people on unrelated social behaviour to their detriment? — that is social scoring, prohibited; does it assess credit standing? — that is high-risk. Does it infer emotions of a worker or a student? — prohibited; of anyone else, for another purpose? — potentially high-risk. Does it predict offending purely from profiling, with no human-led evidence? — prohibited; does it support an officer's assessment with real case evidence? — high-risk. Is it live facial identification in a public square by police? — prohibited (bar narrow exceptions); is it after-the-fact or non-public biometric ID? — high-risk.
Worked example: Veldkamp BV, ~210 employees, ~€41M turnover
Veldkamp, a Rotterdam HR-tech firm, scopes three ideas. (1) A tool that infers employees' stress and engagement from webcam expressions during reviews: emotion recognition in the workplace → Article 5(1)(f) prohibited. Veldkamp kills it. (2) A feature ranking candidates for clients: Annex III point 4 employment → high-risk, allowed with the full stack. (3) A "trustworthiness score" blending unrelated social-media behaviour to flag applicants: this strays toward social scoring (Article 5(1)(c)) detrimental treatment in an unrelated context — Veldkamp drops the social-behaviour signals and confines the feature to job-relevant data, keeping it inside the high-risk lane rather than the prohibited one. One idea banned, one built with controls, one redesigned to stay on the lawful side.
How Confir helps
Confir runs the Article 5 prohibition screen before the Article 6 high-risk check, so a banned use is caught and stopped rather than wrongly worked up as high-risk — and a high-risk use gets the right requirements rather than being over-killed as prohibited. The engine is deterministic and rule-based: the same intended purpose always lands on the same side of the line, with the cited Article 5 sub-paragraph or Annex III point that fired shown in plain English.
Related guides
- How AI risk classification works
- How to classify your AI system
- Prohibited AI practices explained
- Social scoring under the EU AI Act
- Emotion recognition in the workplace
- Real-time biometric identification ban
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 →