Prohibited AI — The Lines Nobody Can Cross

Published 30 March 2026 · 1 min read

Eight AI practices are completely banned under the EU AI Act. No compliance pathway exists. No transition period applies. The prohibition has been in force since 2 February 2025.

The eight prohibited practices

Social scoring by public or private actors. Exploitation of vulnerabilities due to age, disability, or economic situation. Real-time remote biometric identification in publicly accessible spaces by law enforcement (with narrow exceptions requiring prior authorisation by a judicial or independent administrative authority). Emotion recognition in workplaces and educational institutions (except for medical or safety purposes). Biometric categorisation based on sensitive attributes such as race, religion, or sexual orientation. Untargeted scraping of facial images to build recognition databases. Predictive policing based solely on profiling. Subliminal manipulation or deceptive techniques that cause significant harm.

The penalty

The highest in the entire AI Act. Up to €35 million or 7% of global annual turnover — whichever is higher. For comparison, the GDPR maximum is €20 million or 4%.

The boundary between prohibited and high-risk

Narrower than most people think. Emotion recognition is prohibited in workplaces and schools — but permitted in other contexts as high-risk. Biometric identification is prohibited in real-time public settings — but post-event identification by law enforcement is high-risk, not prohibited. The same technology can fall on either side depending on where, when, and how it is used.

What to do if you find a prohibited system

Stop immediately — there is no compliance pathway. Conduct a formal assessment documenting the system and the specific Article 5 category. Seek legal advice if the boundary is unclear. Discontinue and document the cessation. Assess whether a modified, non-prohibited version is permissible.

What’s in the full chapter

Our full chapter covers all eight prohibitions in detail, the exact boundary between prohibited and high-risk for each, a five-step response procedure, and an eight-question self-check.