Glossary
Human in the loop
A design in which a person reviews and approves, edits or rejects certain agent actions before they take effect.
Checks run on our GPUs in Switzerland. Request payloads are not stored.Security
What Human in the loop means
Human in the loop (HITL) means a person has a decision point in the agent’s workflow, typically approving or rejecting an action before it runs. The MCP specification’s section on tools says there SHOULD always be a human in the loop with the ability to deny tool invocations.
HITL is only as good as the attention behind it. If every call needs approval, reviewers stop reading and click through (see approval fatigue). If none do, the human is not really in the loop.
The practical goal is to route the right actions to people. MCP Guard’s approval_policy head predicts auto_approve, require_human or reject, and the ask verdict exists for exactly this: hold the call and ask. Combining it with blast radius lets you keep humans for high-impact, uncertain actions and let routine, reversible ones run.
Related terms
How Human in the loop connects to the rest of the vocabulary.