Last reviewed:
What is human in the loop? Definition and AI Act obligations
Human in the loop refers to a mode of operation in which a person approves, corrects or rejects each decision of an AI system before it takes effect. It differs from human on the loop, where the AI acts alone under the supervision of a human able to intervene, and from human out of the loop, where no human intervenes during operation.
The phrase comes from control engineering and the military; the European Union set it for AI in the ethics guidelines of its High-Level Expert Group (April 2019). That text distinguishes three levels: human in the loop, the capability for human intervention in every decision cycle, while noting this is often neither possible nor desirable; human on the loop, intervention during design and monitoring of operation; human in command, the capability to oversee the system's overall activity and to decide when and how to use it. The European AI Regulation (AI Act) turns this principle into an obligation for high-risk systems. Its Article 14 requires them to be designed so that they can be effectively overseen by natural persons. These must be able to understand the system's capacities and limitations, remain aware of automation bias (the tendency to over-rely on the machine), interpret its outputs, decide not to use it or disregard its output, and stop it. For remote biometric identification, a result must be verified by at least two people. Article 26 requires deployers to assign this oversight to people with the necessary competence, training and authority. Since the Digital Omnibus (Regulation (EU) 2026/1744), these obligations apply from 2 December 2027 for Annex III high-risk systems (recruitment, credit, education) and from 2 August 2028 for Annex I systems.
Concrete example
Illustrative case (fictitious company). A network of temp agencies with 250 employees uses an AI tool that ranks the applications received for each assignment. Recruitment is listed in Annex III of the AI Act: the tool is high-risk. First organisation: recruiters receive the sorted list every morning and call the top ten names. Nobody looks at the rejected candidates: the human is present but controls nothing. Revised organisation: the tool proposes a reasoned ranking, a recruiter also reviews a sample of rejected profiles each week, can change the ranking and reports discrepancies; recruiters are trained on the tool's limits and a manager can suspend its use.
Comparison
| Level | Human role | When to use it | Example |
|---|---|---|---|
| Human in the loop | Approves each decision before execution | Decisions that commit a person, an amount or reputation | Credit refusal, candidate shortlisting, sending a major quote |
| Human on the loop | Monitors operation, can intervene or stop | High volume, low unit stakes, recoverable errors | Sorting incoming mail, first-level support replies |
| Human out of the loop | Intervenes only at design time and in after-the-fact checks | Low-risk, fully reversible tasks | Automatic filing of internal documents |
| Human in command | Decides whether, when and how the system is used | Always, at management level | Committee that authorises, suspends or withdraws an AI tool |
FAQ
What is the definition of human in the loop?
An AI system works "human in the loop" when a person reviews and approves each decision or action before it is executed. The AI proposes, the human decides. It is the highest level of oversight, suited to decisions that commit people or significant amounts.
How is human in the loop translated in French?
It is translated as "humain dans la boucle". The French version of the AI Act speaks of "contrôle humain" (Article 14). The abbreviation HITL is also used.
What is the difference between human in the loop and human on the loop?
In the loop: the human approves each decision before execution. On the loop: the AI acts alone, a human monitors operation and can intervene or stop it. Out of the loop: no human intervention during operation, only at design time. The choice depends on the stakes and volume of decisions.
Is human in the loop mandatory?
The AI Act requires effective human oversight for high-risk AI systems (Article 14), applicable from 2 December 2027 for Annex III systems since the Digital Omnibus. The text does not require approving each decision, but being able to understand, disregard, correct and stop the system. The GDPR (Article 22) also regulates solely automated decisions producing legal effects.
How do you apply human in the loop to AI agents?
You define mandatory approval points: before a message to a customer, a payment, a data deletion or a change to a production system. The agent prepares, the human approves. For the rest, on-the-loop oversight with logging and the ability to stop is often enough.
See also
Further reading
Regulation (EU) 2024/1689 (AI Act), official text on EUR-Lex
Sources
- Regulation (EU) 2024/1689 on artificial intelligence (AI Act), Article 14 "Human oversight" and Article 26, Official Journal of the EU, 12 July 2024. https://eur-lex.europa.eu/eli/reg/2024/1689/oj
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), postponement of high-risk obligations, Official Journal of the EU, 2026. https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX%3A32026R1744
- Ethics Guidelines for Trustworthy AI, High-Level Expert Group on AI (AI HLEG), European Commission, April 2019. https://www.europarl.europa.eu/cmsdata/196377/AI%20HLEG_Ethics%20Guidelines%20for%20Trustworthy%20AI.pdf