What Makes a Practice Prohibited?
Article 5 of Regulation EU 2024/1689 establishes absolute prohibitions, practices so incompatible with EU values and human dignity that no compliance programme, risk management system, or regulatory approval can render them lawful. These are not requirements with which one must comply; they are bans that must be observed by halting the prohibited activity entirely.
The prohibitions became enforceable on 2 February 2025, six months after the Regulation entered into force. Any organisation operating a prohibited AI system since that date has been in continuous violation.
The Nine Prohibited AI Practices
Subliminal Manipulation
AI deploying subliminal or deceptive techniques beyond conscious awareness to influence behaviour in ways that cause or are likely to cause harm. Covers neuromarketing AI exploiting psychological weaknesses, dark-pattern AI that bypasses rational decision-making, and persuasion AI operating below awareness thresholds.
Exploitation of Vulnerabilities
AI targeting specific groups based on age, disability, or socioeconomic situation to distort behaviour in ways likely to cause harm. Includes AI targeting children with manipulative content, systems exploiting elderly cognitive vulnerabilities, and AI targeting economically distressed individuals with predatory offers.
Social Scoring by Public Authorities
Government AI systems evaluating or classifying persons based on social behaviour or personal characteristics, leading to detrimental treatment unrelated to the context of original data collection or disproportionate to the behaviour's gravity. Directly addresses state social credit systems.
Criminal Prediction from Profiling
Law enforcement AI making individual risk assessments for crime prevention based solely on profiling or personality traits, without any objective, verifiable facts directly linked to criminal activity. Does not prohibit AI using specific, concrete behavioural evidence.
Facial Image Scraping for Biometric Databases
AI creating or expanding facial recognition databases through untargeted scraping of facial images from the internet or CCTV footage. Addresses practices that led to regulatory action against companies like Clearview AI across EU member states. Prohibits building biometric databases without specific, targeted data collection with proper legal basis.
Emotion Inference in Workplaces and Education
AI inferring employees' or students' emotional states, except where used for medical or safety reasons. Prohibits emotion-sensing productivity monitoring, AI monitoring student engagement through facial expression analysis, and similar workplace or classroom surveillance technologies that infer emotional state.
Biometric Categorisation for Sensitive Attributes
AI categorising persons based on biometric data to deduce or infer race, political opinions, trade union membership, religious or philosophical beliefs, or sexual orientation. Addresses AI systems inferring protected characteristics from physical appearance, facial geometry, gait analysis, or other biometric signals.
Real-Time Remote Biometric ID in Public Spaces
Live facial recognition and similar systems in publicly accessible spaces by law enforcement. Three narrow exceptions apply: missing children; imminent terrorist threats; serious criminal investigations, all requiring prior judicial or independent administrative authorisation. For non-law-enforcement entities: absolute prohibition with no exceptions whatsoever.
AI-Generated Intimate Imagery (NCII) and CSAM
Applies 2 Dec 2026Added by the Digital Omnibus on AI. Prohibits placing on the market, putting into service, or using AI systems designed to generate or manipulate non-consensual intimate imagery (so-called "nudifier" apps) of an identifiable person, or to generate child sexual abuse material (CSAM). The prohibition binds both providers and deployers, and reaches systems where such generation is a reasonably foreseeable outcome and the system lacks effective technical safeguards to prevent it. A safe harbour applies for systems with adequate preventive safeguards. Unlike the other Article 5 prohibitions (in force since 2 February 2025), this one applies from 2 December 2026. It becomes legally binding once the Omnibus is published in the EU Official Journal.
Article 5 Compliance Checklist
Before deploying any AI system, answer these questions:
- Does this system operate below users' conscious awareness to influence their behaviour?
- Does this system target vulnerable groups, children, elderly, economically distressed, to distort their behaviour?
- Is this system operated by a public authority to evaluate persons based on social behaviour?
- Does this system make criminal predictions based purely on profiling without concrete factual links?
- Does this system build biometric databases through mass scraping of images?
- Does this system infer employee or student emotional states?
- Does this system categorise persons by race, religion, political opinion, or sexual orientation from biometric data?
- Does this system use live facial recognition in public spaces?
A "yes" to any question means the system is operating a prohibited practice in violation of Article 5. Operation must cease immediately.
The only exceptions are for law enforcement use of real-time remote biometric identification under Article 5(1)(h), subject to very narrow conditions (specific criminal offences, imminent threats, missing children) and mandatory prior judicial or independent administrative authorisation. For all non-law-enforcement organisations, the prohibited practices under Article 5(1)(a)-(h) are absolute bans with zero exceptions. The Digital Omnibus adds a ninth prohibition under Article 5(1)(i) banning AI-generated non-consensual intimate imagery and CSAM, which applies from 2 December 2026. Even for law enforcement, the exceptions are strictly bounded and subject to rigorous procedural requirements in each member state. Note that post-hoc (retrospective) biometric identification is not prohibited under Article 5; it is regulated as high-risk under Annex III.