Human in the loop
A design principle where AI proposes and a named person decides, so that every material compliance decision has accountable human approval.
Overview
In compliance, the person who approves a decision is accountable for it to supervisors and courts. AI can read, map, test and draft at a scale no team can, but an AI system cannot carry that accountability.
Human-in-the-loop systems record both the AI’s proposal, including its confidence and reasoning, and the human decision, so an examiner can see exactly where judgement was applied.
It connects most directly to Impact note, Evidence package and EU AI Act.
Check the official sources below for legal scope, applicable dates and exceptions. This entry is an explanatory reference, not legal advice.
Why it matters
It is the difference between AI that reduces compliance risk and AI that creates a new kind of it.
Who it affects
Any organisation using AI in regulated decisions; increasingly a requirement under AI regulation.
Connections
The terms most closely tied to Human in the loop. Select one to open it.
What to do about it
- 1Map where your obligations come from.
- 2Link each one to the controls and evidence behind it.
- 3Review the map whenever the business or the rules change.
From requirement to evidence.
Regulation Layer never closes an issue, grants an exception or changes a control on its own. Every material decision carries a named approver.
Request early access →