security//human-in-the-loop

Human-in-the-loop (HITL) design is a system design in which a person must review or authorize selected parts of a system's trajectory, so that consequential decisions wait for human judgment. The phrase says little until three details are specified: which actions stop, what evidence the reviewer sees, and what occurs after rejection.

Human-in-the-loop (HITL) design is a system design in which a person must review or authorize selected parts of a system's trajectory, so that consequential decisions wait for human judgment. The phrase says little until three details are specified: which actions stop, what evidence the reviewer sees, and what occurs after rejection.

A gate reduces agent autonomy selectively, without supervising every harmless step, and approvals work best for rare, consequential and legible decisions. If prompts are constant, low-information or routinely safe, the human learns to click through them. The interface then preserves ceremony while losing judgment. Human review is strongest as one independent layer of defense in depth, never the final defense against every technical failure.

A person in the loop is not automatically a control. Sometimes it is only latency wearing a badge.