Login Get a Demo
01 Glossary · Governance and risk

What Is an AI Acceptable Use Policy?

The staff-facing rules for AI: what the policy covers, how it differs from the organization’s AI policy, and how to write one people can follow.

The short answer

What is an AI acceptable use policy?

Last updated
Read time
4 min read
From
Cranium AI
Also called
AI AUP, Generative AI usage policy
Share on LinkedIn

An AI acceptable use policy is the document that tells employees and contractors which AI tools they may use, for which tasks, with which data, and what they must check before relying on the output.

It is the staff-facing layer of AI governance, shorter and more specific than the organization’s overall AI policy.

Key takeaways
  • It governs how people use AI. The broader AI policy governs how the organization builds, buys and oversees it.
  • NIST’s Generative AI Profile names acceptable use policies in suggested actions GV‑1.4‑002 and GV‑3.2‑003.
  • EU AI Act Article 4 obliges providers and deployers to support staff AI literacy. It has applied since February 2, 2025.
  • Each rule needs a matching control or record. Otherwise nobody can tell whether it is followed.
02 What it covers

What does an AI acceptable use policy cover?

01

Approved tools

Names the AI tools staff may use, the accounts to use them with, and the route for requesting a new one. Anything not on the list needs a request first.

AI use case intake
02

Data rules

States which classes of data may go into which tools. Typical never-enter items are customer personal data, credentials and secrets, unreleased financials and source code, unless a tool is approved for them.

03

Output rules

Requires a person to check AI output before relying on it, and sets a higher bar, human review and a record, for output that informs decisions about people.

Human oversight
04

Prohibited uses

Lists uses that are off limits whatever the tool, starting with the law. EU AI Act Article 5 bans practices such as social scoring and inferring emotions in the workplace.

Article 5
03 Compare

AI acceptable use policy vs. AI policy

AI acceptable use policyAI policy
AudienceEvery employee and contractorLeadership, risk, legal, engineering and procurement
ScopeDay-to-day use of AI toolsHow AI is built, bought, assessed, overseen and retired
ReferenceNIST AI 600‑1, actions GV‑1.4‑002 and GV‑3.2‑003ISO/IEC 42001 clause 5.2; NIST AI RMF GOVERN 1.2
FormA few pages of plain rules with examplesPrinciples, roles, risk tolerance and required processes
Enforced throughTraining, access controls and monitoringAssessments, approvals, audit and management review
04 How to write

How do you write one?

  1. Start from the inventory

    Write rules for the AI tools people already use, not a hypothetical list. The inventory and shadow AI findings tell you which those are.

  2. Tie data rules to data classes

    Reuse your existing data classification, so staff apply labels they already know.

  3. Name the tools and the request route

    Publish the approved list and a quick way to ask for an addition. A policy with no route to yes produces workarounds.

  4. Train, then record acknowledgment

    Article 4 asks for measures that fit staff knowledge and context of use. It does not require guaranteeing any specific level of literacy.

  5. Review when the tools change

    NIST action GV‑6.1‑010 suggests updating acceptable use policies to address proprietary and open-source generative AI technologies and data.

05 Keep going

Related reading.

06 Questions

Frequently asked questions.

Straight answers, kept current. Last updated .

Ask us anything →
Is an AI acceptable use policy legally required?

No law requires a document with that title. The EU AI Act does require providers and deployers to take measures to support staff AI literacy (Article 4), and requires deployers of high-risk systems to use them in line with the instructions for use (Article 26(1)). A written policy is the usual way to show both. ISO/IEC 42001 requires an AI policy in clause 5.2.

What should employees never enter into a public AI tool?

It depends on the tool’s terms and your data classification, so the policy should answer by data class. Common entries on the never list are personal data about customers and employees, passwords, keys and other secrets, unreleased financial results, source code, and anything covered by a confidentiality agreement.

How do you enforce an AI acceptable use policy?

Pair every rule with a control or a record. Training and signed acknowledgment show that staff know the rules. Single sign-on limits access to approved tools. Monitoring for unapproved AI use shows where the policy is being bypassed, and a fast request route removes the reason to bypass it. Review the findings on a set schedule.

07 See it for yourself

From the definition
to the evidence.

See how Cranium AI discovers, observes, governs, secures and proves every AI model, agent and vendor in your enterprise.