Automation bias is the tendency of people to trust the output of an automated system more than their own judgement or conflicting evidence. It produces two kinds of error: a wrong recommendation is accepted, or a problem goes unnoticed because the system raised no alert. The effect grows with time pressure, routine and a long run of correct results.
The AI Act uses the term explicitly: persons who oversee a high-risk AI system must be put in a position to stay aware of this tendency (Art. 14(4)(b)). Useful countermeasures are spot checks with independent review, a duty to give reasons for approvals, training on the limits of the system and interfaces that make uncertainty visible. The threat T03 Excessive automation (automation bias) describes causes and suitable controls.