Human-in-the-loop approval is a control that requires a person to review and confirm a sensitive action before an AI system, script, or operator tool can execute it. The goal is simple: keep automation from making irreversible decisions on its own. In security work, this is often used for actions such as sending data externally, deleting resources, changing permissions, running privileged commands, or accessing production systems.
This matters because AI agents can act quickly and chain together many steps, which makes mistakes or misuse more damaging. A human approval gate adds a final check for intent, scope, and safety, helping block prompt injection, overbroad permissions, and misconfigured test systems from causing real harm. In defenses, it is paired with least privilege, logging, and egress controls so that high-impact actions require explicit review instead of silent execution.