In a human-in-the-loop setup, the AI handles defined tasks, such as scoring CVs against criteria, running a first screening interview or ranking a shortlist, and passes its output to a person. The recruiter can see the reasoning, review the evidence, change scores and decide who progresses or is declined. In a well-designed setup, no outcome reaches a candidate until a person has approved it.
Keeping people in charge of decisions catches mistakes a tool cannot see, keeps accountability with the employer and supports compliance with rules on automated decision-making, such as GDPR and the EU AI Act. The oversight has to be real: reviewers need time, training, access to the evidence and genuine authority to disagree with the tool, rather than rubber-stamping its suggestions.