Unacceptable-Risk AI: Article 5 Ban

Unacceptable-risk AI covers a shortlist of practices that the EU considers so harmful they are simply forbidden, not just regulated. Think of it as the red line: if an AI system uses subliminal techniques to manipulate behaviour, exploits vulnerabilities of a specific group (like age or disability), enables social scoring by public authorities, or involves real-time remote biometric identification in public spaces for law enforcement (with very narrow exceptions), it is banned outright. The key contrast is with high-risk AI, which is allowed but heavily regulated—so never confuse a high-risk system with a prohibited one.

To avoid traps, remember the mnemonic 'MESS' for the four main categories: Manipulation, Exploitation, Social scoring, and Surveillance (real-time biometric). A common exam trick is to describe a high-risk scenario, like an AI for credit scoring, and ask if it is prohibited—it is not, because only social scoring by public authorities is banned. Another trick: if a question mentions 'emotion recognition in the workplace', that is high-risk, not prohibited, unless it is used to manipulate. Always check whether the practice is on the Article 5 list; if it is not, it is either high-risk, limited-risk, or minimal-risk.

To lock this in, test yourself with a quick scenario: 'An AI system ranks citizens based on their social media activity to allocate welfare benefits.' That is social scoring by a public authority, so it is prohibited. If the ranking is done by a private bank for loan decisions, it is high-risk, not banned. Remember: if it is not in Article 5, it is not prohibited—just regulated.