Human-in-the-Loop (HITL)
Human-in-the-loop is a design pattern in which a human reviews, approves, or overrides an AI system's decision before it is executed. In agent governance, HITL is the mechanism that ensures high-risk actions receive human judgment rather than being auto-approved by policy.
Why It Matters
The EU AI Act explicitly requires human oversight for high-risk AI systems. HITL is the operational implementation of that requirement: a person evaluates the agent's proposed action and decides whether it should proceed.
The challenge is calibrating when HITL applies. Too much human involvement (every action requires approval) creates consent fatigue and negates the productivity benefits of agents. Too little (everything is auto-approved) creates governance gaps.
Effective HITL systems use policy to route: low-risk actions are auto-approved, high-risk actions are escalated to a human, and prohibited actions are denied outright. The human sees only the decisions that genuinely require judgment. Suspend-and-resume is the mechanism that makes HITL practical for multi-step agent tasks.