Trust metrics are measurements that show whether people rely on an AI system in a sensible, consistent way. They can include override rates, acceptance rates, time spent reviewing outputs, manual workarounds, and how often teams use the system in normal workflows. In cyber security, these metrics matter because an AI tool that is technically accurate can still fail if analysts ignore it, while a tool that is trusted too much can hide mistakes or false confidence.
Defenders use trust metrics to spot overreliance, underuse, and uneven adoption across teams. For example, if analysts always accept model alerts without review, that may signal blind trust. If they constantly bypass the system, that may show low confidence or poor workflow fit. Attackers can benefit from either problem: overtrust may let malicious activity slip through, while undertrust can create gaps and inconsistent controls. Tracking trust helps security teams tune thresholds, training, and human oversight so AI supports decisions without taking them over.