An AI policy tells your teams three things: which data they may give to an AI, which tools they are allowed to use, and who reviews a text prepared by an AI before it goes to a client. One or two pages are enough, as long as they are kept up to date.
This template is a starting point, not legal advice. Have your policy reviewed by the person in charge of your personal data, or by your lawyer.
What should an AI policy contain?
- The data you may give to an AI.
- Texts with no names and no customer data: a letter outline, an idea for an article, a general question.
- The data you do not give to an unauthorised tool.
- Personal data of customers and employees, contracts, company figures, passwords. The GDPR requires you to know where personal data goes and who processes it.
- The authorised tools.
- A short list, with for each one: what it is for, who may use it, and with which account. A personal account is not a company tool.
- Human approval.
- Nothing an AI prepares goes to a client, a supplier or an authority without someone having read it. You must be able to find out who approved it.
- Training.
- The EU AI Act requires companies that use AI to make sure their staff know how to use it (Article 4).
- A point of contact.
- Someone to ask a question or report a mistake to.
What does an AI policy template look like?
A seven-point structure, to reuse and adapt to your company.
Purpose
Why the policy exists, and who it applies to: employees, interns, contractors.
Authorised tools
The list of AI tools allowed for work, with the account to use. Other tools are not used for work.
Allowed and forbidden data
What may be given to each tool, and what is never given: customer and employee data, contracts, figures, passwords.
Review and approval
Any text prepared by an AI is read before it is sent. The person who approves it is accountable for it.
Checking sources
An AI can make mistakes or make things up. A figure, a date or a clause is checked in the original document.
Training
Everyone learns to ask, review and approve before using the tool.
Point of contact and updates
The name of the point of contact, the date of the policy, and a review with each new tool.
How does Kap help you follow the policy?
A policy is easier to follow when the tool points the same way.
| Criterion | Kap | ChatGPT or Copilot |
|---|---|---|
| Where your data is processed | In your offices with the Kap Box, or in Europe with Kap Cloud | At the provider |
| Human approval, with a record | Yes, for every message sent | Not built in |
| Source cited in every answer | Yes | Depends on the question |
| Prepares the work before you ask | Yes | No, it answers when asked |
Official sources.
Official texts consulted on 16 November 2026. This page helps you ask the right questions: it does not replace advice from your data protection officer or your lawyer.
- EU AI Act, Regulation (EU) 2024/1689, Article 4 (EUR-Lex, French text) : https://eur-lex.europa.eu/legal-content/FR/TXT/HTML/?uri=CELEX:32024R1689
- GDPR, Regulation (EU) 2016/679 (EUR-Lex, French text) : https://eur-lex.europa.eu/legal-content/FR/TXT/HTML/?uri=CELEX:32016R0679
- CNIL (French data protection authority), “AI: how to comply with the GDPR?” (in French) : https://www.cnil.fr/fr/intelligence-artificielle/ia-comment-etre-en-conformite-avec-le-rgpd
Read next.
The questions we are asked.
What is a company AI policy?
An AI policy tells your teams three things: which data they may give to an AI, which tools they are allowed to use, and who reviews a text prepared by an AI before it goes to a client. One or two pages are enough, as long as they are kept up to date.
What should go into an AI policy?
The authorised tools, the data you may give and the data you never give, review before sending, checking sources, training, and the name of a point of contact.
Is an AI policy mandatory?
Neither the GDPR nor the EU AI Act uses the word “policy” in this sense. But the first requires you to control where personal data goes, and the second to make sure your teams know how to use AI. A policy is the simplest way to organise this.
Can an employee use their personal ChatGPT account for work?
That is for your policy to say. With an online assistant, what you write is processed by the provider, and the terms depend on the plan and the settings. A contract pasted into a personal account therefore leaves the company.
How does Kap help you follow an AI policy?
Kap runs in a space of your own hosted in Europe, or on a device in your offices: nothing is sent to an external AI provider. Nothing goes out without human approval, and each approval is recorded with the person’s name and the time.
AI sometimes makes things up. How can you trust its answers?
Kap answers from your documents and cites its source every time: you check it in one click. And nothing goes out until a person has approved it.
See Kap on your own files.
The first Kap call is free and takes one hour, by video call.