An AI acceptable use policy template gives employees clear rules for using AI at work: which tools are approved, what data can go into them, what is off limits, when a person must check the output, and how to report a problem. The AI acceptable use policy template below covers all of that in ten short sections, plus one most policies leave out: AI agents and coding assistants that run commands and connect to MCP servers with an employee's own access. Copy it, fill in the bracketed parts, and have legal and HR review it. It is a starting point, not legal advice.
What an AI acceptable use policy template should cover
A useful policy answers the questions employees actually have: can I use this tool, can I paste this document into it, and who do I ask. It should be short enough to read in five minutes, specific enough to act on, and tied to a list of approved tools that someone keeps current.
NIST's AI Risk Management Framework, released on January 26, 2023 for voluntary use, organizes AI risk work into four functions: govern, map, measure and manage. An acceptable use policy is part of the govern function. NIST's Generative AI Profile, published on July 26, 2024, applies the same framework to generative AI and is a good reference if you need to justify a rule to leadership.
The AI acceptable use policy template
1. Purpose and scope
This policy sets the rules for using artificial intelligence tools at [Company]. It applies to all employees, contractors and anyone else using [Company] accounts, devices or data, whether the tool is provided by [Company] or not.
2. Approved tools
Use only the AI tools on the approved list at [link], with your [Company] account. To request a new tool, submit [form or ticket]. Personal AI accounts may not be used with [Company] data.
3. Data you may and may not use
Public information may be used with any approved tool. Internal information may be used only with tools marked "internal data allowed". Confidential information, customer data, personal data, credentials, source code marked restricted, and anything covered by a contract or regulation may not be entered into any AI tool unless the approved list says that tool is cleared for it.
4. Prohibited uses
Do not use AI to: make final decisions about hiring, firing, pay, credit or other matters that significantly affect a person; create content that is deceptive, discriminatory or harassing; impersonate a real person; produce work that breaks a law, a contract or a third party's rights; or get around a security control.
5. Human review
You are responsible for anything you publish, send, ship or decide using AI output. Check facts, figures, citations and code before using them. Label AI-generated content where [Company] or a customer requires it.
6. AI agents and coding assistants
AI agents that can run commands, change files, call APIs or act on your behalf may be used only if they are on the approved list. Do not give an agent production credentials, administrator tokens or access to customer data unless [security team] has approved that setup. Keep the agent's approval prompts turned on for commands that delete, deploy or send data outside [Company].
7. MCP servers, plugins and integrations
MCP servers, plugins, extensions and other tools that connect an AI agent to systems or data must be approved before use, including ones added to a project's configuration file. Install them only from the publisher named on the approved list, with a pinned version. Do not put secrets directly in configuration files.
8. Accounts and security
Sign in to AI tools with single sign-on where available. Do not share AI accounts or API keys. Keep data retention and training settings at the values [Company] sets.
9. Reporting problems
Report to [security contact] right away if you entered data that this policy does not allow, an AI tool or agent did something you did not ask for, or you suspect a tool or MCP server is malicious. Reporting quickly is never a policy breach.
10. Enforcement and review
[Company] may monitor the use of AI tools on its accounts and devices, as described in [monitoring policy]. Breaking this policy may lead to loss of access and disciplinary action. [Owner] reviews this policy at least every six months.
How to adapt the AI acceptable use policy template
Start with the approved tools list, because the rest of the policy points to it. For each tool, record which data classes it is cleared for, based on its contract, data retention and training terms. Then match section 3 to the data classes you already use; if you have no classification scheme, public, internal and confidential is enough to begin.
Regulated companies should add their specific obligations to section 3, such as health, payment or student data, and name the regulation. Companies with works councils or unions should check section 10 with HR before announcing monitoring. Keep the language plain: a rule nobody understands is a rule nobody follows.
Decide who owns the approved tools list before you publish. In most companies that is security or IT, with legal signing off on data terms. Give the owner a target for answering requests, such as five working days, and publish it. The fastest way to create shadow AI is a policy that says "ask first" and a request queue nobody answers.
The section most policies miss: AI agents and MCP servers
Most AI policies were written for chat tools, where the risk is what an employee types in. Coding agents add a second risk: what the agent does. It can run shell commands, install packages and load MCP servers that read files, query databases and call outside services, all with the employee's own access.
The OWASP Top 10 for Agentic Applications, published in December 2025, lists identity and privilege abuse and agentic supply chain vulnerabilities among its top risks, and the OWASP MCP Top 10 names shadow MCP servers as a risk on their own. The postmark-mcp case shows why section 7 matters: a fake email server on npm, installed like any other tool, copied every email it sent to an attacker. Requiring approval for MCP servers is how a policy covers that.
Making the policy stick
A policy works only if people know it and it matches reality. Announce it with the approved tools list, add it to onboarding, and give a named person who answers questions within a day. Make requesting a tool easy, or people will skip the request.
Then check it against what is actually in use. Our guide on how to find shadow AI in your organization covers SaaS and network checks, and our guide to finding MCP servers on developer machines covers the coding agents the policy now includes. Run those checks before you announce the policy and again each quarter. If the first check finds tools nobody approved, treat it as a request queue rather than a list of violations: approve what is safe, offer an alternative for what is not, and give people a date to switch. The MCP security checklist gives reviewers a consistent way to approve new servers under section 7.
Next step: check your AI acceptable use policy template against reality
An AI acceptable use policy template only protects you if it matches the tools people really run. Run a Discovery Scan to list every AI agent and MCP server across your machines and repositories, and see where practice and policy differ.