Human in the Loop (HITL)

Human in the loop (HITL) keeps a person in the decision path of an automated system, typically to review, approve, or correct the AI before a consequential action proceeds. It is the pragmatic answer to the fact that models are neither perfectly accurate nor accountable: a human takes responsibility where the stakes are too high to automate outright.

In a governed system, human review is an outcome policy can require, not a manual afterthought. A request that exceeds a spend cap, touches sensitive data, or falls below a confidence threshold can be escalated to a reviewer instead of being allowed or silently blocked - which is also how obligations like the EU AI Act's human-oversight requirement become an enforced control.