Holders can choose an appropriate AI tool for a task and place it within the EU AI Act's risk categories. They check output critically rather than accepting it, recognise hallucination and over-reliance, and know when to adjust, escalate or stop.