Skip to content
Search
AppliedGlossary

AI Acceptable Use Policy

Many people already use AI at work, approved or not (shadow AI). A short policy turns that into something you can manage. It should answer five questions in plain words: which tools are approved and on which accounts; which kinds of data may never go into them; which outputs need a person's review before use (human in the loop); how AI use is disclosed to clients where it matters; and who to ask about a tool or task the policy doesn't cover.

Keep it to one or two pages, give examples for each rule, and make the approved tool easy to get, or people will route around the policy. Review it a few times a year, because the tools change quickly. For a larger organization, the NIST AI Risk Management Framework gives a fuller structure for governing AI risk. For a small business, a clear data rule backed by PII redaction and a named person to ask covers most of the risk. Guardrails in your own AI features can enforce parts of it automatically.

Example

AI acceptable use policy (outline)
1. Approved tools: [tool names], company accounts only.
2. Never enter: client financials, health information, passwords, unreleased contracts. Remove names and contact details first.
3. Always review before use: anything sent to a client, published, or used for a hiring, pay or legal decision.
4. Disclosure: tell a client when AI drafted a deliverable if they ask, or if their contract requires it.
5. Questions and new tools: ask [name] before you sign up for anything new.
Last reviewed: [date]

When to use it

  • Any company where staff use AI tools, which is most of them
  • After finding unapproved AI tools in use
  • Before rolling out an AI assistant to a team

When not to use it

  • Copying a long template nobody will read
  • Writing a ban with no approved alternative