Prohibited AI practices
Article 5 lists AI practices the EU considers an unacceptable threat to people's safety, livelihoods and rights. These bans have applied since 2 February 2025 and carry the Act's highest fines.
Learning objectives
- Identify each prohibited practice in Article 5
- Explain the narrow exceptions, especially for biometric identification
- Recognise the new prohibitions added in 2026
- Spot prohibited practices hidden inside everyday products
3.1 The prohibited practices
It is prohibited to place on the market, put into service or use AI systems that do the following:
1. Harmful manipulation and deception
Using subliminal techniques beyond a person's awareness, or purposefully manipulative or deceptive techniques, that materially distort behaviour by impairing people's ability to make an informed decision — causing, or reasonably likely to cause, significant harm.
2. Exploiting vulnerabilities
Exploiting 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.
A voice-enabled toy that encourages children to take dangerous dares; a lending app that targets people in financial distress with manipulative prompts to take high-cost loans.
3. Social scoring
Evaluating or classifying people over time based on their social behaviour or personal characteristics, where the resulting score leads to detrimental treatment that is either unrelated to the context in which the data was collected, or unjustified or disproportionate. This applies to both public and private actors.
4. Predictive policing based solely on profiling
Assessing the risk that a person will commit a crime based solely on profiling or personality traits. AI that supports a human assessment already based on objective, verifiable facts directly linked to criminal activity is not covered.
5. Untargeted scraping of facial images
Creating or expanding facial recognition databases through untargeted scraping of facial images from the internet or CCTV footage.
6. Emotion recognition at work and in education
Inferring the emotions of people in the workplace or in educational institutions — except where the system is intended for medical or safety reasons.
Prohibited: a call-centre tool that rates agents' "enthusiasm" from their voice. Allowed: a system that detects driver drowsiness for safety.
7. Biometric categorisation of sensitive traits
Categorising individuals based on biometric data to deduce or infer their race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation. Labelling or filtering of lawfully acquired biometric datasets, for example in law enforcement, is excepted.
8. Real-time remote biometric identification for law enforcement
Using real-time remote biometric identification (such as live facial recognition) in publicly accessible spaces for law enforcement — unless strictly necessary for one of three objectives:
- a targeted search for specific victims of abduction, trafficking or sexual exploitation, or for missing persons;
- preventing a specific, substantial and imminent threat to life or physical safety, or a genuine threat of a terrorist attack;
- locating or identifying a suspect of certain serious crimes (listed in Annex II and punishable by at least four years' detention).
Even then, use needs prior authorisation from a judicial or independent administrative authority (with an urgency procedure), a fundamental rights impact assessment, registration, and must be permitted by national law.
The Digital Omnibus on AI added bans on AI systems that generate or manipulate:
- realistic non-consensual intimate imagery of identifiable people — so-called "nudifier" apps — covering images, video and audio; and
- child sexual abuse material (CSAM).
These apply from 2 December 2026, and also cover systems where such output is a reasonably foreseeable result and reasonable safeguards are missing.
3.2 What the prohibitions mean in practice
- No risk management "fix". Unlike high-risk AI, there is no compliance route — a prohibited system must not be offered or used.
- Providers and deployers are both responsible. A deployer using a lawful tool in a prohibited way (for example, using a general sentiment tool to monitor employees' emotions) breaches the Act.
- Highest penalties: up to €35 million or 7% of total worldwide annual turnover, whichever is higher.
- Guidance: the European Commission published guidelines on prohibited practices in February 2025, with practical examples — worth reading alongside the Act.
An HR tech vendor adds a "wellbeing dashboard" to its video-meeting plug-in. It analyses facial expressions in meetings and flags employees who appear "disengaged" or "stressed" to their managers.
Analysis: this is emotion recognition in the workplace, and it isn't for medical or safety purposes. It is prohibited — regardless of the "wellbeing" framing. The vendor shouldn't offer it in the EU, and employers shouldn't use it.
Key takeaways
- Article 5 bans manipulation, exploitation of vulnerabilities, social scoring, profiling-only crime prediction, facial-image scraping, workplace and education emotion recognition, sensitive biometric categorisation and (with narrow exceptions) real-time remote biometric ID for law enforcement.
- Two new bans — on nudifier and CSAM-generating AI — apply from 2 December 2026.
- Prohibitions have applied since 2 February 2025 and carry fines of up to €35m or 7% of global turnover.