High-risk AI: the listed-sector test

The AI Act defines high-risk AI by where it is used, not just what it does. If an AI system operates in a listed sensitive sector—such as employment (hiring software), education (exam grading), law enforcement (facial recognition), migration (visa processing), or essential services (credit scoring)—or acts as a safety component of a product (like a braking system in a car), it is automatically high-risk. The key contrast: a medical diagnosis AI used in a hospital is high-risk because healthcare is a listed sector, but the same AI used in a fitness app for general wellness is not, even if it gives wrong advice.

To spot high-risk systems fast, remember the mnemonic 'EELMES' for the sectors: Employment, Education, Law enforcement, Migration, Essential services, Safety components. A common trick in questions is to describe a system that feels dangerous—like an AI that predicts loan defaults—but if it is not in a listed sector (e.g., used by a private club for membership decisions), it is not high-risk. Another trap: a system that is a safety component, like an AI in a toy’s voice recognition, is high-risk only if the toy itself is regulated under product safety laws; otherwise, it falls outside the scope.

To test your understanding, ask yourself: 'Is this AI in one of the six EELMES categories?' If yes, it is high-risk; if no, it is not, regardless of how harmful it might seem. For the exam, remember that the Act’s list is exhaustive—no sector outside those six triggers high-risk status, so focus on matching the scenario to the list, not on your intuition about risk.