Human in the loop
HITL · Human on the loop · Human-in-the-loop AI
Human in the loop is a setup in which a person must approve an AI system's action before it takes effect. Human on the loop means that the system acts and a person watches over it, with the means to intervene. Human out of the loop means nobody intervenes during execution.
The right person at the right point
Putting a human everywhere is neither possible nor desirable. The useful principle is a person at the right decision point, with a real capacity to intervene, to stop and to audit, and the heavier a decision is and the harder to undo, the higher it should go. In ACF the setting depends on the zone of the decision: green decisions run with a person on the loop, orange decisions with a person in the loop, and red decisions remain with the person.
Three quantities that break the loop
When any of these crosses a threshold, the rule is no longer being applied even though the procedure still exists on paper. Such is the point at which a control applied by software before execution, such as ACF Control, takes over from the person, and the person then deals with true exceptions.
| Quantity | What happens |
|---|---|
| Volume | Validation turns into blind signature |
| Speed | There is no time left to escalate before the action takes effect |
| Irreversibility | A payment sent, an account closed or a statement published cannot be taken back afterward |
What makes human review real
- The reviewer has the information needed to refuse, and enough time to use it.
- Refusing has no cost for the reviewer.
- Approval rates are monitored: a rate of 100 percent over months should raise a question.
- A customer can ask for a person at any time, by a means that does not pass through the agent.
Frequently asked questions
Does the AI Act require a human in the loop?
It requires effective human oversight for high-risk systems, in proportion to risk and autonomy, without imposing approval of each action. One specific case does require verification by at least two people: decisions based on remote biometric identification.
When is human in the loop the wrong choice?
When volume or speed make the review meaningless. A nominal review is worse than none, because it gives false assurance and lets a person take the blame for a decision they did not really take.
Where does your organization stand?
ACF Score measures your agentic governance in about 10 minutes, the AI Act checker tells you which obligations apply, and ACF Control applies your rules before an agent acts.
Related terms
Primary sources
Written by Vincent Dorange, creator of the Agentic Commerce Framework®. Last updated .