Human in the Loop — AI Governance Guidelines | Tablecloth

AI Governance Guidelines

Human in the Loop — what it means and what it asks of you

A practical reference for anyone using AI tools at work. Covers oversight levels, what to check before approving AI output, and when to escalate.

Free Resource AI Governance Committee
Start here

Human in the Loop means a person stays actively involved when using AI — reviewing, checking, or approving what the AI produces before it's used or acted on.

One rule that matters: Approving an AI output means you own it. "The AI wrote it" is not a defense if something goes wrong.

Oversight Framework

The two levels of oversight

Not every AI task carries the same risk. Understanding which level applies keeps accountability clear and processes consistent.

Level 1

Review & Approve

AI produces the output, but a human must review and sign off before it is used, sent, or acted on.

Typical examples

  • Legal documents
  • External communications
  • Reports and marketing copy
  • Code before deployment
Level 2

Automated with Oversight

AI runs autonomously. Humans review outcomes regularly and can escalate at any time.

Typical examples

  • Email sorting
  • Scheduling
  • Internal lookups

Before You Approve

What to check in any AI output

These checks apply any time you're signing off on something an AI produced. They're not bureaucratic steps — they're how you protect yourself and the organization.

Facts & Sources

AI can state incorrect information confidently. Verify factual claims, statistics, and cited sources independently. Do not assume a source is real or accurate just because the AI mentioned it.

Math & Figures

AI frequently makes errors with calculations, percentages, and data. Any numbers in an AI output should be checked manually against the original data before use.

Completeness

Has the AI missed anything important? Outputs can be plausible but incomplete. Consider what's not there, not just what is.

Appropriateness

Is the content suitable? Check for bias, unintended implications, or anything that could be misread by its intended audience.

Decision Guide

When to apply more oversight

Work through these questions before using any AI output. Follow the path — each answer determines the next step.

START Evaluate the AI / Data Task Does this involve personal, sensitive, or regulated data? YES Review & Approve NO Will this be sent externally, published, or executed? YES Review & Approve NO Could this affect someone's job, finances, health, or reputation? YES Review & Approve NO Am I using this tool in a way it wasn't originally approved? YES Stop and check with the committee NO Is this routine, reversible, and low-stakes? YES Automated with Oversight NO Review & Approve
When in doubt, apply more oversight, not less. The cost of over-reviewing is low. The cost of approving something that goes wrong is not.

Your Role

Your responsibilities as an employee

Using AI tools responsibly means active engagement — not passive approval. These are the baseline expectations for everyone in the organization.

Read AI outputs properly before approving — don't rubber-stamp.

Check facts, figures, and claims where the stakes are high.

Report outputs that seem wrong, biased, or unusual — don't just fix and move on.

Don't use approved tools for tasks outside their approved scope without checking first.

Ready to go further

Build an AI governance structure that actually holds.

Tablecloth works with organizations to put the right oversight, accountability, and tooling in place — so AI adoption doesn't outpace your controls.

Whether you're looking for a structured program or a quick conversation about where to start, we have a path for you.