Prohibited AI practices sit at the very top of the EU AI Act’s risk pyramid. Set out in Article 5, they are the uses of AI considered to pose an unacceptable risk — so harmful to fundamental rights, safety, or democratic values that they are banned outright rather than merely regulated. Unlike high-risk systems, which may be placed on the market once they meet stringent requirements, prohibited practices cannot be made lawful by any amount of documentation, oversight, or safeguarding. They were also among the first provisions of the Act to take effect.
Why some practices are banned outright
The logic of an outright ban is that for certain uses, the potential harm is so severe or so contrary to European values that no risk-mitigation measures could justify them. Where the high-risk regime says “you may do this if you do it carefully”, the prohibitions say “you may not do this at all”. This makes Article 5 categorically different from the rest of the Act, and it makes confirming that none of an organisation’s AI falls within it the very first compliance question to settle.
Manipulative and deceptive techniques
The Act prohibits AI systems that deploy subliminal techniques beyond a person’s awareness, or purposefully manipulative or deceptive techniques, with the effect of materially distorting behaviour in a way that causes or is likely to cause significant harm. The concern is AI that bypasses rational decision-making to push people toward choices they would not otherwise make. The threshold involves both distortion of behaviour and a link to significant harm, but the prohibition reflects a clear principle: AI should not be used to covertly manipulate people.
Exploiting vulnerabilities
A related prohibition targets AI systems that exploit the vulnerabilities of a person or group — due to age, disability, or a specific social or economic situation — with the effect of materially distorting their behaviour in a way that causes or is likely to cause significant harm. This protects those least able to protect themselves from being singled out by AI designed to take advantage of their circumstances. It reflects a recurring theme of the Act: heightened protection where power between the system and the individual is most unequal.
Social scoring
The Act prohibits social scoring — AI systems that evaluate or classify people over time based on their social behaviour or personal characteristics, where the resulting score leads to detrimental treatment that is unjustified, disproportionate, or unrelated to the context in which the data was originally generated. The provision guards against the emergence of pervasive scoring systems that follow people across unrelated areas of life and penalise them, a scenario widely seen as incompatible with European fundamental rights.
Biometric prohibitions
Several prohibitions concern biometrics. The Act bans, with narrow and tightly defined exceptions, the use of real-time remote biometric identification in publicly accessible spaces for law-enforcement purposes. It also restricts certain biometric categorisation systems that infer sensitive attributes, the untargeted scraping of facial images to build recognition databases, and emotion recognition in the workplace and educational settings. These provisions reflect particular sensitivity around technologies that can identify or profile people at scale without their knowledge or consent.
The narrow exceptions
Where the Act permits otherwise-banned biometric uses — chiefly certain real-time identification for specific, serious law-enforcement objectives — it surrounds them with strict conditions and safeguards. These exceptions are deliberately narrow and subject to authorisation and oversight; they are not a general licence. For ordinary organisations outside the law-enforcement context, the practical takeaway is simpler: the biometric prohibitions apply, and the exceptions are unlikely to be available.
Why early confirmation matters
Because the prohibitions took effect early and because no compliance measures can rescue a banned practice, confirming that none of an organisation’s AI falls within Article 5 is logically the first step in any AI Act programme. There is no point assessing risk tiers or preparing documentation for a system that is simply not allowed. A quick but careful screen against the prohibited practices should therefore precede the rest of the classification work.
How prohibitions differ from high-risk
It is worth being precise about the distinction. A high-risk system is permitted subject to obligations; a prohibited practice is not permitted at all. The two are sometimes confused because both involve serious concerns, but the legal consequences are entirely different. Mistaking a prohibited practice for a merely high-risk one — and investing in compliance documentation for something that can never be lawful — is a costly error that a clear understanding of Article 5 prevents.
What organisations should do
The practical approach is a focused screen: review each AI use against the categories in Article 5 — manipulation, exploitation of vulnerabilities, social scoring, and the biometric prohibitions — and document the conclusion. For the vast majority of business uses, the answer will be that no prohibited practice is involved, and the assessment moves on to risk tiers. Where there is any doubt, the use should be examined closely before deployment. Innopulse builds this Article 5 screen into the first stage of its AI Risk Check, so prohibited uses are caught before any further effort is spent.
Conclusion
Prohibited AI practices are the EU AI Act’s absolute limits: uses — manipulation, exploitation of vulnerabilities, social scoring, and certain biometric applications — banned outright under Article 5 because they pose an unacceptable risk. No safeguard makes them lawful, and they took effect early in the Act’s timeline. Confirming that none of an organisation’s AI falls within these categories is the logical first step of compliance, cleanly separating what must simply stop from what may continue under the high-risk regime.