AI Acceptable Use Policy: What to Put in It (Template)
The sections every company AI acceptable use policy needs, from approved tools and data rules to custom agents and enforcement, plus a template to adapt.
Most companies already have AI in daily use. Some of it runs on a tool the company pays for and manages. A lot of it runs on personal ChatGPT, Claude or Gemini accounts that sit outside the company's admin controls and retention policies. An AI acceptable use policy is how you draw the line between the two, and tell people clearly what they can do instead of only what they can't.
This guide covers what to put in the policy and how to make it stick. It isn't legal advice; have your counsel review the final text, especially if you handle regulated data.
Start with the approved tool, not the ban
Policies that only ban AI tend to push the use further out of sight. People keep using AI, just on their phones and personal accounts, where you can see even less. A policy works when it names an approved tool that is at least as convenient as the one people are reaching for.
For a company on Microsoft 365, that tool is usually Microsoft Copilot. Copilot Chat comes with Microsoft 365 business subscriptions at no extra cost, it works inside the apps people already use, and enterprise data protection applies when people sign in with their work account. The paid Microsoft Copilot license adds Copilot inside Word, Excel, PowerPoint, Outlook and Teams with company data, and the policy doesn't need to change for that.
The sections to include
1. Purpose and scope
Say who the policy covers (employees, contractors, anyone working with company data) and what counts as AI: chat assistants, AI features inside other apps, browser extensions, and custom assistants or agents.
2. Approved tools
Name the approved tools and how people get access. Be specific about accounts: "Microsoft Copilot, signed in with your company account". The same tool used with a personal account is a different product with different protections. Say how to request a new tool, and who decides.
3. Data rules
This is the heart of the policy. Use the data categories you already have, and state for each one where it may go:
| Data | Approved company AI tool | Personal or unapproved AI tools |
|---|---|---|
| Public information | Allowed | Allowed |
| Internal information | Allowed | Not allowed |
| Confidential and client data | Allowed, following existing access rules | Not allowed |
| Regulated data (health, payment, personal data) | Only where your compliance team approves | Not allowed |
If you use Microsoft Purview sensitivity labels, refer to them by name. Copilot respects them: content it creates from labeled sources inherits the highest-priority label.
4. Personal AI accounts
State plainly that company and client data doesn't go into personal AI accounts, and why: those accounts are outside the company's admin controls, retention policies and contracts.
5. Checking the output
People stay responsible for what they send, publish or decide. The policy should ask them to check facts, numbers and citations before use, and to follow any disclosure rules your industry or clients set.
6. Custom assistants and agents
Custom GPTs, Claude Projects, Gems and Copilot agents are small applications, and they need owners. Cover who may build them, who may share them and with whom, where their knowledge files live, and who reviews an agent before it's shared widely. In Microsoft Copilot, Agent Builder agents are shared with specific people, and publishing an agent across the organization goes through Copilot Studio with admin approval, which gives you a natural review point.
7. Monitoring and enforcement
Tell people what you monitor, then do it with the tools you already have. On Microsoft 365:
- Find the AI tools in use. Microsoft Defender for Cloud Apps lists generative AI apps in its cloud app catalog.
- Block unapproved apps. Apps marked unsanctioned are blocked on devices onboarded to Defender for Endpoint.
- See what data reaches AI sites. Microsoft Purview DSPM for AI can show visits and sensitive data sent to third-party AI sites such as ChatGPT and Gemini when the Purview browser extension is deployed.
- Stop sensitive data at the browser. Endpoint DLP can block pasting or uploading sensitive data to AI sites.
Start in simulation or audit mode, look at what people actually do, and switch on blocking once the approved tool is in their hands.
8. Training and review
Give people a short session on the approved tool and the data rules, not just a document to sign. Review the policy on a fixed schedule; AI tools and vendor terms change quickly.
A one-page outline to start from
- Purpose and scope
- Approved AI tools and how to get access
- What data may go where (the table above)
- Personal AI accounts
- Responsibility for output
- Custom assistants and agents: building, sharing, ownership
- Monitoring and enforcement
- Requests for new tools
- Training and review schedule
Template you can adapt
Replace the bracketed parts, cut what doesn't apply, and have your counsel review it before it goes out.
1. Purpose and scope. This policy explains how [Company] employees, contractors and anyone working with [Company] data may use artificial intelligence tools, including chat assistants, AI features inside other software, browser extensions, and custom assistants or agents.
2. Approved tools. The approved AI tool is [Microsoft Copilot], used only while signed in with your [Company] account. Other AI tools may be used for [Company] work only with written approval from [IT / the AI owner]. To request a tool, contact [name or address].
3. Data. Public information may be used with any AI tool. Internal, confidential and client information may be used only in approved tools, and only within your existing access rights. [Regulated data such as health, payment or personal data] may be used only where [the compliance team] has approved it in writing.
4. Personal accounts. Do not enter [Company] or client information into personal AI accounts or free consumer AI tools.
5. Responsibility. You are responsible for anything you send, publish or decide using AI output. Check facts, figures, names and citations before use, and follow any disclosure rules set by clients or regulators.
6. Custom assistants and agents. Only [roles] may build agents. Agents may be shared with named colleagues; sharing with the whole company requires review by [owner]. Every agent has a named owner, and its knowledge files are stored in [the team's SharePoint library].
7. Monitoring. [Company] monitors the use of AI tools on company devices and accounts to protect company and client data, and may block unapproved tools.
8. Training and review. Everyone completes [Company]'s AI training before using approved tools. [Owner] reviews this policy every [six months].
9. Questions and breaches. Ask [contact] if you are unsure whether a use is allowed. Report accidental sharing of confidential data to [contact] right away.
Policy is half the job
A policy without an approved tool people like is a document nobody follows. The other half is giving them that tool and moving their existing work into it: the custom GPTs, the saved prompts, the Projects and Gems their teams depend on. Our shadow AI guide covers how to find what's in use, and the AI Migration Assessment inventories every AI tool your people use, sanctioned or not, and plans the move to Microsoft Copilot.