The EU AI Act bans eight AI practices under Article 5, including harmful manipulation, social scoring, emotion recognition at work and in schools, untargeted facial-image scraping and most real-time biometric identification by police. These bans have applied since 2 February 2025. Two more, on non-consensual intimate imagery and child sexual abuse material, apply from 2 December 2026.
Key takeaways
- Article 5 is the "unacceptable risk" tier: these uses cannot be made compliant, only stopped.
- The original eight prohibitions have applied since 2 February 2025, and fines for breaching them since 2 August 2025.
- The Digital Omnibus, Regulation (EU) 2026/1744, adds bans on AI that generates non-consensual intimate imagery and child sexual abuse material from 2 December 2026.
- Several bans carry narrow exceptions, for example emotion recognition at work is allowed for medical or safety reasons.
- Breaches carry the Act's highest fines: up to €35 million or 7% of worldwide annual turnover.
Why Article 5 matters even if you never build AI
Most companies assume the bans target authoritarian states or fringe start-ups. In practice the risk sits in ordinary procurement. An HR analytics suite with a "team mood" feature that reads faces on video calls, a sales tool promising to detect a customer's emotional state, a loyalty programme that scores customers on behaviour unrelated to their purchases: each could fall inside Article 5, and the ban applies to the organisation that uses the system, not only to the vendor that built it.
This is the top tier of the Act's four-level structure; the other tiers are covered in EU AI Act risk classification explained, and the wider framework in our EU AI Act guide. If you are unsure whether any tool in your estate goes near these lines, a free AI Act gap assessment from Opsio asks the screening questions in a structured order and gives you a written AI Acceptable Use Policy that can rule these practices out internally.
Need help with cloud?
Book a free 30-minute meeting with one of our cloud specialists. We'll analyse your situation and provide actionable recommendations — no obligation, no cost.
The eight prohibitions in force since 2 February 2025
The legal text is in Article 5 of the AI Act. The table below paraphrases each point and gives a plain-English example. Most points ban placing such a system on the market, putting it into service and using it.
| Art. 5(1) | What is banned | Plain-English example | Key exception |
|---|---|---|---|
| (a) | Subliminal, manipulative or deceptive techniques that materially distort behaviour and cause, or are likely to cause, significant harm | An app that uses hidden cues to push vulnerable users into gambling far beyond their means | None; lawful persuasion such as normal advertising is not caught unless it meets the harm threshold |
| (b) | Exploiting vulnerabilities due to age, disability or a specific social or economic situation, causing significant harm | A toy voice assistant that nudges children into risky behaviour; targeting people in debt with predatory loan offers | None |
| (c) | Social scoring that leads to detrimental treatment in unrelated contexts or out of proportion to the behaviour | A landlord platform that rejects tenants based on a score built from their social media activity | Applies to public and private actors alike; lawful, purpose-specific scoring such as credit scoring is regulated elsewhere |
| (d) | Predicting the risk of a person committing a crime based solely on profiling or personality traits | A tool that flags individuals as likely shoplifters from demographic and personality data | Allowed to support a human assessment already based on objective, verifiable facts linked to criminal activity |
| (e) | Creating or expanding facial recognition databases through untargeted scraping of facial images from the internet or CCTV | A vendor that harvests social media photos to build a face-search service | None |
| (f) | Inferring the emotions of people in the workplace or in education institutions | Webcam software that scores employee engagement from facial expressions during meetings | Allowed for medical or safety reasons, such as detecting driver fatigue |
| (g) | Biometric categorisation to deduce race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation | A camera system that sorts shoppers by inferred ethnicity or religion | Labelling or filtering of lawfully acquired biometric datasets, and categorisation in law enforcement |
| (h) | Real-time remote biometric identification in publicly accessible spaces for law enforcement | Live facial recognition on a city's street cameras matched against a police database | Strictly necessary searches for victims or missing persons, preventing an imminent threat to life or a terrorist attack, and locating suspects of serious crimes, subject to prior authorisation and other safeguards |
How to read the harm threshold in points (a) and (b)
Points (a) and (b) are the ones most likely to touch mainstream products, and they are also the most qualified. Three elements need to be present: a manipulative, deceptive or exploitative technique; a material distortion of the person's behaviour, impairing their ability to make an informed decision; and significant harm, or a reasonable likelihood of it, to that person or others. A recommendation engine that shows you more of what you like does not meet that bar. A design that deliberately exploits a known addiction to keep someone spending well beyond their means may.
Emotion recognition: banned at work, regulated elsewhere
Point (f) is narrower than many assume. It applies only in workplaces and education institutions. A call centre system that infers customer emotions is not banned by point (f), but it is a high-risk use under Annex III and triggers transparency duties towards the people it analyses. Inside the workplace, the medical and safety exception is real but tight. Fatigue detection for drivers or pilots qualifies; "wellbeing" monitoring used to evaluate staff performance does not. Classification of the permitted cases is covered in our article on high-risk AI systems and Annex III.
Real-time biometric identification by police
Point (h) bans live remote biometric identification in public spaces for law enforcement, with three exceptions. Even then, Article 5(2) to (7) require prior authorisation by a judicial or independent administrative authority, a fundamental rights impact assessment, registration of the system, and national legislation that permits the use. Private security firms are outside point (h) but not outside the Act: their remote biometric identification systems are high-risk under Annex III point 1.
Two new prohibitions from 2 December 2026
The Digital Omnibus on AI (Regulation (EU) 2026/1744), in force since 27 July 2026, added two points to Article 5(1). Both apply from 2 December 2026.
| New prohibition | What it covers | Who is caught |
|---|---|---|
| Non-consensual intimate imagery | AI that generates or manipulates realistic images, video or audio depicting an identifiable person's intimate parts or sexual activity without their freely given, specific, informed, unambiguous and explicit consent ("nudifier" apps) | Providers where this is the intended purpose, or where the design makes it a reasonably foreseeable outcome without adequate safeguards; deployers who use a system to produce such material |
| Child sexual abuse material | AI that generates or manipulates child sexual abuse material as defined in Directive 2011/93/EU | Same provider and deployer logic as above |
For providers of general image and video generators, the practical question is safeguards. A model that can be prompted into producing this content without effective filters risks falling under the ban even if that was never its purpose. Expect authorities to look at prompt filtering, output classifiers, abuse reporting and how quickly the provider acts on misuse. The wider set of changes is explained in what the Digital Omnibus changed in the AI Act.
The Commission guidelines on prohibited practices
On 4 February 2025 the European Commission published its guidelines on prohibited AI practices, a long document that works through each point with examples of what is and is not caught. They are not legally binding; only the Court of Justice of the European Union can give an authoritative interpretation. They are still the best available indication of how the Commission and national authorities read Article 5, and worth reading for anyone building or buying systems near these lines. Note that they predate the two Omnibus prohibitions.
Enforcement and fines
Breaching Article 5 can lead to administrative fines of up to €35 million or 7% of total worldwide annual turnover for the preceding financial year, whichever is higher. For SMEs and start-ups the lower of the two figures applies. National market surveillance authorities enforce the bans, and the penalty provisions have applied since 2 August 2025. The detail, including how fines are set, is in EU AI Act penalties and fines.
How to screen your organisation for prohibited practices
- Inventory every AI capability, including features inside HR, CRM, security and video-conferencing tools. Emotion and biometric features are often switched on by default.
- Run each one against the eight points plus the two new ones. Ask what the system infers, about whom, and in what setting.
- Check the exceptions narrowly. If you rely on the medical or safety exception, record why, and who decided.
- Write the bans into policy. An AI acceptable use policy should name the prohibited uses so staff do not adopt them through a free trial.
- Add Article 5 to vendor due diligence, with contractual warranties that features will not be used or enabled for prohibited purposes.
- Repeat when tools change. A vendor update can add an emotion-analytics module overnight.
Dates for the remaining obligations, which now stretch to 2028, are set out in our EU AI Act timeline and deadlines.
Frequently asked questions
What is banned under the EU AI Act?
Article 5 bans harmful manipulation, exploitation of vulnerable groups, social scoring, crime prediction based solely on profiling, untargeted scraping of facial images, emotion recognition at work and in schools (save for medical or safety reasons), biometric categorisation by sensitive traits, and most real-time biometric identification by police in public spaces. From 2 December 2026, AI that generates non-consensual intimate imagery or child sexual abuse material is also banned.
What is an unacceptable use of AI?
An unacceptable use is one the AI Act treats as a clear threat to safety, livelihoods or fundamental rights, so it is prohibited rather than regulated. These are the practices listed in Article 5. Unlike high-risk systems, they cannot be made lawful through documentation, testing or human oversight; the only compliant response is not to place them on the market, put them into service or use them.
When did the AI Act prohibitions take effect?
The original eight prohibitions have applied since 2 February 2025, six months after the AI Act entered into force on 1 August 2024. Fines for breaching them became possible from 2 August 2025, when the penalties chapter started to apply. The two prohibitions added by Regulation (EU) 2026/1744 apply from 2 December 2026.
Can we use AI to analyse employee emotions or engagement?
Not if the system infers emotions of people in the workplace, unless it is intended for medical or safety reasons, such as detecting fatigue in drivers. Engagement scores based on facial expressions or voice tone during meetings fall inside the ban. Analysing anonymous survey text for themes is a different activity, but check whether it infers emotions of identifiable individuals.
Does Article 5 apply to companies outside the EU?
Yes. The AI Act applies to providers that place AI systems on the EU market regardless of where they are established, and to providers and deployers outside the EU where the system's output is used in the EU. A US or UK vendor selling an emotion-recognition feature to EU employers is caught by the ban.
Rule out prohibited practices in writing
The cheapest way to avoid an Article 5 breach is to make sure no one in your organisation adopts a banned tool in the first place. Get your AI acceptable use policy written from your answers with Opsio's free EU AI Act readiness questionnaire: about 40 minutes, four editable Word documents (AI Governance Framework, AI Acceptable Use Policy, AI Vendor Due Diligence Checklist, AI Risk Assessment Template) and a maturity score across ten governance areas. Free, with no card and no obligation.
This article is general information, not legal advice. Check the consolidated text of Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744. Statements are accurate as of October 2026.
Related Guides
Written By
Country Manager, India
Praveena leads Opsio's India operations, bringing 17+ years of cross-industry experience spanning AI, manufacturing, DevOps, and managed services.
Editorial standards: This article was written by cloud practitioners and reviewed by our engineering team. We revise it when the technology or regulation it covers changes; the dates on this page show the latest update. Opsio maintains editorial independence.