Risk & governance
Human in the loop
A person reviews, approves or corrects AI-assisted work.
Also known as: HITL
Definition
Human-in-the-loop design assigns defined decisions or exceptions to authorised people before or after an AI system acts.
Why it matters
It is only a real control when reviewers have sufficient evidence, time, authority and a usable escalation path.
Related concepts
- Guardrail
A technical or procedural control that constrains system behaviour.
- Audit trail
A durable record of inputs, evidence, decisions, actions and versions.
- Fallback
A safe alternative when the preferred AI path cannot continue.