What AI practices are prohibited under the EU AI Act?

Use the statutory definitions

The original prohibitions began applying on 2 February 2025. Social scoring is not confined to public authorities. Remote biometric identification is different from biometric categorisation; the earlier version of this page conflated them. The workplace emotion-inference prohibition has medical or safety exceptions, not a general permission based on employee consent. Retrospective identification is subject to its own requirements and should not be described as universally prohibited.

Check new and existing practices separately

The Commission states that the Omnibus added a prohibition covering AI systems generating non-consensual sexually explicit or intimate content or child sexual abuse material, applicable from December 2026. That later date does not postpone the original bans. If a practice appears to fall within a prohibition, suspend the affected use while obtaining a specific legal assessment. Document the feature, purpose and data involved so the assessment addresses the actual practice rather than a general product description.

Reviewed 4 October 2026. Sources: Commission AI Act overview and timetable, classification, deployer duties, literacy guidance and AI Office of Ireland.