This page covers the prohibited practices and the high-risk rules from our EU AI Act guide, which also explains the four risk tiers, the implementation timeline and what to do now. Transparency, GPAI models and fines are in Article 50 transparency, GPAI models and fines.
Prohibited AI practices
The original Article 5 prohibitions have applied since 2 February 2025. Each one has a precise legal test and, in some cases, exceptions, so check a tool against the wording of the Article rather than the summary below.
- Manipulative or deceptive techniques: AI that uses subliminal, manipulative or deceptive techniques to materially distort behaviour in a way that causes, or is reasonably likely to cause, significant harm.
- Exploiting vulnerabilities: AI that exploits vulnerabilities linked to age, disability or a specific social or economic situation to materially distort behaviour and cause significant harm.
- Social scoring: AI that evaluates or classifies people or groups based on their social behaviour or personal characteristics, where the resulting score leads to detrimental treatment in unrelated contexts, or treatment that is unjustified or disproportionate. The prohibition is not limited to public authorities; it applies to private organisations too.
- Criminal risk prediction by profiling: AI that assesses the risk of a person committing a crime based solely on profiling or personality traits, rather than objective, verifiable facts linked to criminal activity.
- Untargeted facial image scraping: AI that creates or expands facial recognition databases through untargeted scraping of images from the internet or CCTV footage.
- Emotion recognition at work and in education: AI that infers the emotions of people in the workplace or in education institutions, except where it is used for medical or safety reasons.
- Biometric categorisation by sensitive attributes: AI that categorises individuals from biometric data to deduce race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation, with a limited exception for lawfully acquired datasets in law enforcement.
- Real-time remote biometric identification for law enforcement: live identification in publicly accessible spaces for law enforcement purposes, permitted only for narrowly defined objectives under prior authorisation and safeguards. This is a law enforcement rule, not a general ban on facial recognition.
New prohibitions from 2 December 2026
The Digital Omnibus added two practices to Article 5, applying from 2 December 2026. The first covers AI systems that generate or manipulate realistic images, video, audio or similar content depicting an identifiable person in intimate parts or sexually explicit activity without their consent, which is aimed at so-called nudification tools. The second covers AI used to generate child sexual abuse material. Both come with specific definitions and exclusions in the legal text, so a provider of an image or video generator should review them against its own product before December.
Employee engagement platforms, productivity tools that read facial expressions or tone of voice, and HR technology that categorises workers by biometric signals may fall within the emotion recognition or biometric categorisation prohibitions. Review those tools against the exact Article 5 wording.
High-risk AI: Annex III and Annex I
There are two routes into the high-risk category. Annex III lists areas of use, and the Digital Omnibus moved the obligations for those systems to 2 December 2027. Through Annex I, Article 6(1) covers AI systems that are themselves products, or safety components of products, covered by the EU harmonisation legislation listed in Annex I, such as machinery or medical devices, where that product must also undergo a third-party conformity assessment; those obligations now apply from 2 August 2028. The original August 2026 and August 2027 dates no longer apply. Classification depends on the intended purpose and the legal conditions, not simply on the sector in which the AI is used.
Listing in Annex III does not settle the classification on its own. Under Article 6(3), an Annex III system is not high-risk if it does not pose a significant risk of harm and does not materially influence the outcome of decisions, for example where it performs a narrow procedural task, improves the result of work a person has already completed, detects patterns in decisions without replacing human assessment, or performs a preparatory task. The provider must document that assessment. An Annex III system that profiles natural persons is always high-risk.
A worked example: a CV-screening tool that ranks or filters applicants and materially influences who is invited to interview is high-risk under Annex III (employment). An administrative tool that only summarises documents a recruiter has already selected, without ranking candidates or steering the decision, may fall under the Article 6(3) exception, provided the provider documents that assessment. If either tool profiles the applicants, the exception does not apply and the system is high-risk.
What deployers of high-risk AI must do
If your organisation uses a high-risk AI system, your obligations under Article 26 are separate from the provider's. The provider handles design, technical documentation and conformity assessment. You are responsible for how the system is used.
- Use it according to the instructions: take appropriate technical and organisational measures to use the system in line with the provider's instructions for use. Changing the system or its purpose can change your legal position. Under Article 25, a deployer is treated as the provider of a high-risk system, with the provider obligations that follow, if it puts its name or trademark on that system (without prejudice to contracts that allocate the obligations differently), makes a substantial modification so that it remains high-risk, or changes the intended purpose of a system so that it becomes high-risk. Each change needs a case-by-case assessment; ordinary configuration within the provider's instructions does not normally have this effect.
- Assign human oversight: give oversight to people with the necessary competence, training, authority and support. Article 14 requires high-risk systems to be designed for effective oversight in proportion to the risk. That does not mean every output needs manual approval, but your people must be able to monitor, intervene and stop the system where needed. Data protection law adds its own rules on decisions based solely on automated processing.
- Monitor and report: monitor the system's operation and inform the provider and, where required, the market surveillance authority of risks and serious incidents.
- Keep the logs: keep the logs the system generates automatically, to the extent they are under your control, for a period appropriate to the intended purpose and for at least six months, unless other EU or national law provides otherwise (Article 26(6)). This is separate from the provider's ten-year documentation retention.
- Inform workers: before putting a high-risk system into use at the workplace, inform the workers' representatives and the affected workers.
- Fundamental rights impact assessment where required: Article 27 requires a FRIA from bodies governed by public law, private entities providing public services, and deployers of systems for creditworthiness assessment and credit scoring, or for risk assessment and pricing in life and health insurance. It does not apply to every private deployer of high-risk AI. Where personal data is involved, coordinate it with the data protection impact assessment (DPIA) required under the GDPR.
What to ask a provider of high-risk AI
Before buying or deploying a high-risk system, ask for the instructions for use, the known limitations, the human oversight measures, the logs the system produces and the provider's plan for meeting obligations that apply from December 2027. Depending on the timetable and the type of system, check the EU declaration of conformity, the CE marking and any registration requirement. Not all technical documentation is meant for customers, and registration rules vary by type of system and purpose, so focus on what you need to use the system lawfully.
2027
2028