What is human in the loop?

Definition

Human in the loop means a person checks AI work at the points that matter: before a summary goes in the file, before an email goes to a client, before an agent takes an action. The AI does the first draft; a person decides.

FINRA's 2026 oversight report mentions human-in-the-loop review of model outputs as part of monitoring, and suggests firms consider where to have human-in-the-loop oversight for AI agents.

Why it matters for advisors

It's the single most important control for generative AI in an advisory firm, because it catches hallucinations, tone problems and compliance issues before they reach a client. Most AI policies make it a rule for anything client-facing.

It also needs to be real. A reviewer who approves 200 summaries a day without reading them isn't in the loop.

Design the review so it's easy to do well: show the source next to the AI output, flag the parts the reviewer should check closely (numbers, names, recommendations), and keep a record of the approval in the same system the work lives in.

Example

A firm's policy says: AI may draft client emails and meeting recaps, but the advisor must read and approve each one, and the CRM records who approved it.

Watch out for

Write down which outputs need review and by whom. "Someone should check it" isn't a control.