Last updated: 6 November 2026
Quick answer: Article 5 of the EU AI Act bans unacceptable-risk AI, including harmful manipulation, exploiting vulnerabilities, social scoring, profiling-based crime prediction, untargeted facial scraping, emotion recognition at work or school, sensitive biometric categorisation and most real-time biometric identification by police. The bans have applied since 2 February 2025, with fines up to EUR 35 million or 7% of turnover.
The EU AI Act prohibited practices are the first part of the regulation that became binding. Unlike the high-risk rules, which have been pushed back, the Article 5 bans are already in force, and national authorities can fine breaches.
The list also grew in 2026. The AI Omnibus added two new prohibitions on AI-generated intimate imagery and child sexual abuse material, applying from 2 December 2026. This guide explains every prohibited practice, the exceptions, who is liable and how to screen your own AI use.
What are the EU AI Act prohibited practices?
They are uses of AI that the EU considers incompatible with fundamental rights, listed in Article 5 of Regulation (EU) 2024/1689. They cannot be made compliant through documentation or risk controls. If your system falls within a prohibition, you must stop placing it on the market, putting it into service or using it.
Article 5 applies to providers, who develop or market the system, and to deployers, who use it in a professional capacity. Under Article 113, Chapters I and II, which contain the prohibitions, applied from 2 February 2025.
What does each prohibition cover?
There are eight original prohibitions in Article 5(1), points (a) to (h). Each has specific conditions, so read them as legal tests rather than headlines.
(a) Subliminal, manipulative or deceptive techniques
AI that deploys subliminal techniques beyond a person’s consciousness, or purposefully manipulative or deceptive techniques, with the objective or effect of materially distorting behaviour by impairing informed decision-making, in a way that causes or is reasonably likely to cause significant harm.
(b) Exploiting vulnerabilities
AI that exploits 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.
(c) Social scoring
Evaluating or classifying people over time based on social behaviour or personal characteristics, where the score leads to detrimental treatment in unrelated contexts, or treatment that is unjustified or disproportionate to the behaviour.
(d) Predicting criminal offences based solely on profiling
Assessing the risk that a person will commit a crime based solely on profiling or on personality traits. The ban does not cover AI that supports a human assessment already based on objective and verifiable facts directly linked to criminal activity.
(e) Untargeted scraping of facial images
Creating or expanding facial recognition databases through untargeted scraping of facial images from the internet or CCTV footage.
(f) Emotion recognition at work and in education
Inferring emotions of people in the workplace or in education institutions, except where the system is intended for medical or safety reasons.
(g) Biometric categorisation of sensitive traits
Categorising individuals on the basis of biometric data to deduce or infer race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation. Labelling or filtering of lawfully acquired biometric datasets, and categorisation in the area of law enforcement, are carved out.
(h) Real-time remote biometric identification for law enforcement
Using real-time remote biometric identification in publicly accessible spaces for law enforcement, unless strictly necessary for one of three objectives: searching for specific victims of abduction, trafficking or sexual exploitation and missing persons; preventing a specific, substantial and imminent threat to life or a genuine threat of a terrorist attack; or locating suspects of certain serious crimes. Each use needs prior authorisation by a judicial authority or an independent administrative authority whose decision is binding, and the exception is only available where a Member State has provided for it in national law.
New in 2026: intimate imagery and CSAM
Regulation (EU) 2026/1744, the AI Omnibus, entered into force on 27 July 2026 and inserted two new points into Article 5(1):
- Point (ba): AI systems that generate or manipulate realistic images, video or audio depicting an identifiable person’s intimate parts or sexually explicit activity without that person’s explicit consent.
- Point (bb): AI systems that generate or manipulate child sexual abuse material.
These two prohibitions apply from 2 December 2026. Obligations differ between providers and deployers: providers are caught where generating such content is an intended purpose or a reasonably foreseeable outcome without adequate safeguards, while deployers are caught when they deliberately use a system for this purpose.
Which prohibitions matter most for businesses?
For most private organisations, the realistic exposure sits in points (a), (b), (c), (f) and (g), and from December 2026 in (ba) for any business offering image or video generation.
| Prohibition | Where it can appear in business | Key exception or limit |
|---|---|---|
| (a) Manipulation | Dark-pattern chatbots, persuasive recommender tuning, sales agents | Requires material distortion and significant harm |
| (b) Vulnerabilities | Targeting minors, older users or financially distressed customers | Requires significant harm |
| (c) Social scoring | Cross-context customer or tenant scoring | Legitimate, context-specific scoring is not banned |
| (f) Emotion recognition | Interview analysis, call centre sentiment on staff, proctoring | Medical or safety purposes |
| (g) Biometric categorisation | Inferring sensitive traits from faces or voices | Dataset labelling and filtering |
| (ba) Intimate imagery | Image and video generation tools | Explicit consent; applies from 2 December 2026 |
Common mistake: Assuming emotion recognition means only facial analysis. Tools that infer staff emotions from voice, gestures or other biometric signals in a workplace setting can also fall within point (f). Check how the vendor describes the output, not just the input.
What are the penalties for prohibited AI practices?
Breaching Article 5 carries the highest fines in the AI Act. Under Article 99(3), fines reach up to EUR 35 million or, for undertakings, up to 7% of total worldwide annual turnover for the preceding financial year, whichever is higher.
For SMEs and start-ups, Article 99(6) caps each fine at the lower of the two amounts. Under Article 113, the penalty provisions applied from 2 August 2025, so the prohibitions are now fully enforceable by national market surveillance authorities.
The same conduct may also breach the GDPR, for example unlawful processing of biometric data. That means two regulators and two fining regimes can be involved.
How should you interpret the prohibitions?
Use the European Commission’s guidelines. On 4 February 2025 the Commission published guidelines on prohibited AI practices, with examples and explanations of each point. They are non-binding, and the Court of Justice of the EU has the final say on interpretation, but authorities are likely to follow them.
In practice: Record which paragraphs of the guidelines you relied on when you concluded that a use case is not prohibited. If a regulator asks, a reasoned, dated assessment is far stronger than a verbal assurance from a vendor.
How do you screen your AI systems against Article 5?
Screening means inventorying every AI system you build or use and testing each against the prohibitions before looking at high-risk rules. Here is a practical five-step process.
- Build an AI inventory. Include internal tools, vendor products with AI features and general-purpose AI used by staff. Record purpose, users, affected people and data types.
- Identify your role. For each system, decide whether you are the provider, the deployer or both. The obligations and the (ba) test differ by role.
- Run the Article 5 screen. Use the questions below for every system.
- Decide and document. Stop, redesign or confirm the system is outside scope. Record the reasoning and the evidence.
- Control changes. Re-screen when the purpose, user group or model changes, and when you onboard a new AI vendor.
Article 5 screening questions
- Could the system influence people in ways they are not aware of, or in ways designed to mislead, with a risk of significant harm?
- Does it target or disproportionately affect children, older people, people with disabilities or people in financial difficulty?
- Does it score people on behaviour or characteristics and use that score in a different context?
- Does it predict criminal behaviour based only on profiling or personality traits?
- Does it collect facial images at scale from the internet or CCTV?
- Does it infer emotions of employees, candidates or students?
- Does it infer sensitive traits such as ethnicity, religion or sexual orientation from biometric data?
- Can it generate realistic intimate imagery of identifiable people or any sexual content involving minors?
A “yes” to any question needs a documented legal review. Tools such as compliance assessments make this screen repeatable, and where personal data is involved it can sit alongside your data protection impact assessments.
If you want to run the screen across your AI inventory in one workspace, you can start your 14-day free trial and use the AI Governance module.
How do the prohibitions fit with the rest of AI governance?
Article 5 is the first gate in AI governance, not the whole process. Once a system passes the screen, you still need to classify it for high-risk obligations, which the AI Omnibus moved to 2 December 2027 for Annex III systems and 2 August 2028 for Annex I product systems, and to check transparency duties.
Your vendors matter too. Ask AI suppliers to confirm in writing that their systems do not fall within Article 5 and to notify you of changes in intended purpose. A structured vendor and third-party risk management process keeps those confirmations current.
A management system standard such as ISO/IEC 42001 gives you a structure for this ongoing work. Our ISO 42001 certification roadmap explains how to build one, and our post on the overlap between DORA, NIS2 and the EU AI Act covers regulated sectors.
Key takeaways
- The EU AI Act prohibited practices in Article 5 have applied since 2 February 2025, and penalties since 2 August 2025.
- Eight original prohibitions cover manipulation, exploiting vulnerabilities, social scoring, predictive policing by profiling, facial scraping, emotion recognition at work and school, sensitive biometric categorisation and real-time biometric identification by police.
- Two new bans on AI-generated intimate imagery and CSAM apply from 2 December 2026.
- Fines reach EUR 35 million or 7% of worldwide annual turnover, whichever is higher.
- Screen every AI system and vendor against Article 5 and document your reasoning.
Frequently asked questions
When did the EU AI Act prohibitions start to apply?
The prohibitions in Article 5 have applied since 2 February 2025, together with the general provisions in Chapter I. The penalty provisions in Article 99 applied from 2 August 2025. Two additional prohibitions on AI-generated non-consensual intimate imagery and child sexual abuse material, added by the AI Omnibus, apply from 2 December 2026.
Is all emotion recognition banned under the AI Act?
No. Article 5(1)(f) bans AI that infers emotions of people in the workplace and in education institutions, except for medical or safety reasons. Emotion recognition in other contexts is not prohibited, but it is generally classified as high-risk under Annex III and also carries transparency duties toward the people exposed to it.
Does the AI Act ban facial recognition?
Not entirely. It bans building facial recognition databases through untargeted scraping, and bans real-time remote biometric identification in public spaces for law enforcement except in three narrow situations with prior authorisation. Other biometric identification uses are not banned outright, but many are classed as high-risk and must meet strict requirements.
Who is liable for a prohibited AI practice, the provider or the deployer?
Both can be. Article 5 prohibits placing on the market, putting into service and using the listed systems, so providers and deployers each have obligations. Buying a tool from a vendor does not transfer your liability as a deployer, which is why vendor due diligence and your own screening of each use case matter.
What did the AI Omnibus change in Article 5?
Regulation (EU) 2026/1744, in force since 27 July 2026, added two new prohibited practices: AI systems generating or manipulating realistic intimate imagery of identifiable people without explicit consent, and AI systems generating child sexual abuse material. Both apply from 2 December 2026. The new points were inserted alongside the original eight prohibitions rather than replacing them.
Staying on the right side of Article 5
The EU AI Act prohibited practices are a narrow list, but the consequences of getting them wrong are the most serious in the regulation. Inventory your AI use, screen every system and vendor, and document why each use case is permitted. Then move on to high-risk classification and transparency with a clean baseline.
Enactia’s AI-powered GRC platform helps you run AI inventories, Article 5 screens and ISO 42001 controls alongside your GDPR and security work. To see how, contact us or book a demo or start your 14-day free trial.
This article is for general information and is not legal advice.
