Human-in-the-loop

Human-in-the-loop is a system design in which a person reviews, approves or can override an automated system's output before it takes effect, and the person's decision is recorded. It sits on a spectrum: at one end a human approves every consequential action; further along, a human monitors a running system and can intervene, a pattern sometimes called human-on-the-loop.

EU law gives the idea legal shape in two places. Article 14 of the EU AI Act requires high-risk AI systems to be designed so that natural persons can oversee them effectively while they are in use, including the ability to disregard, override or reverse an output and to stop the system, and it asks overseers to stay aware of automation bias, the tendency to over-rely on machine output. Article 26 requires deployers to assign that oversight to people with the necessary competence, training and authority. For stand-alone Annex III systems these duties now apply from 2 December 2027, after the Digital Omnibus on AI moved the date. Separately, Article 22 of the GDPR gives people the right not to be subject to a decision based solely on automated processing that produces legal effects concerning them or similarly significantly affects them, subject to exceptions.

Oversight counts only if it is real. A reviewer who clears hundreds of items an hour, or who cannot see the reasoning and the source behind an output, is a rubber stamp. Workable designs put the gate where errors are costly, show the reviewer the evidence (the source passage, the dates and the system's written reasoning), require a reason for overriding or dismissing, and keep the decision in a tamper-evident audit log. In regulatory monitoring the usual split is that agents collect and propose, and people decide what applies, who owns it and when it is done, as the article on AI agents for regulatory compliance sets out.

This entry is general information, not legal advice.