Human in the loop: keeping a person in charge of AI
2026-11-10
Definition
Human in the loop is a design where an AI system proposes or prepares an action and a person reviews, edits or approves it before it takes effect.
When to use it
Use it wherever a mistake would cost money, upset a customer or break a rule. An AI agent that drafts emails, codes invoices or changes records should usually stop at a draft until someone confirms. This directly limits the damage from hallucination. As trust grows, you can loosen the checks on low-risk steps and keep them on high-risk ones. Writing down which steps need approval is part of sound AI governance.
The review step must be quick and visible. If approving is slow or buried, people click through without reading and the safeguard becomes decoration. See our guide to human review for AI output for a practical setup.
A hypothetical example
Imagine a conveyancing firm that uses AI to prepare the first draft of a contract summary. The AI highlights unusual clauses. A solicitor reads the summary alongside the contract, corrects it and signs off before anything reaches the client.
Deciding where checkpoints belong is a standard part of AI strategy.