An HR manager has Claude score 40 job applications against the role's criteria. A colleague suggests automatically rejecting the 25 lowest-scored applicants to save time. What is the most appropriate approach?

Answer: A — Treat the scores as one input, check them for bias, and have a person make each decision

Rejecting candidates is a decision about people and is hard to reverse, so a human must decide, with the AI output as support and checked for bias (a). Consistency doesn't mean fairness or accuracy (b). A larger model doesn't remove the need for human judgment on hiring (c). A plausible explanation from the model still leaves the decision with the model rather than a person (d).