Unacceptable (prohibited)Real-time biometric surveillance in public spaces, social scoring by governments, subliminal manipulation, exploiting vulnerable groups.

High riskSafety-critical (medical devices, critical infrastructure, autonomous vehicles) and high-impact systems (credit, recruitment, education, law enforcement, border control).

Limited riskTransparency duties, e.g. chatbots and deepfakes: people must know they deal with AI.

Minimal riskRecommender systems, spam filters, games: little or no mandatory obligations.