AI acceptable use policy for your company: a one-page template
- Updated
- By
- The MagnolIA team
- On this page
- 8 min read
Short answer
An AI acceptable use policy is the document that tells staff which AI tools they may use, what information must not go into them and who checks the results. It fits on one page. This MagnolIA template has ten clauses and is a starting model only: adapt it with your data protection officer or legal adviser.
What is an AI acceptable use policy for?
So that nobody has to guess. Once AI becomes part of everyday work, everyone ends up asking the same question: "can I paste this in here?" A short policy answers it before someone pastes a client's contract into a public chatbot.
Two EU laws give reasons to write one, although neither calls it that. Article 4 of the AI Act (applicable since 2 February 2025, according to its application timeline) asks organisations that use AI systems to take measures supporting their staff's AI literacy; since the 2026 omnibus it makes clear that this does not mean guaranteeing a specific level for each person. And Article 32 of the GDPR asks for technical and organisational measures appropriate to the risk whenever personal data is processed.
There is a useful public precedent for the structure. Spain's data protection authority, the AEPD, published its own internal policy on generative AI on 27 November 2025, describing it as the first of its kind in the Spanish public sector. It is limited to the authority's own administrative processes and, according to its press note, is not a way of applying the AI Act. It covers use cases, how tools are selected, the handling of personal data and human oversight.
What should never be pasted into a public AI tool?
Anything that identifies a person, and anything confidential to the business. The list below starts from the data types that MagnolIA's anonymiser detects, and adds three categories worth naming. The identifiers are Spanish because the detector was built for Spanish paperwork; swap in your own country's equivalents.
| What | Examples | Why |
|---|---|---|
| Official identifiers | National ID numbers (in Spain, the DNI and NIE), company tax numbers (CIF) and social security numbers. | They identify a person or a company beyond doubt. |
| Bank details | IBANs and card numbers. | They can be used to move money or impersonate someone. |
| Contact details | Phone numbers, email addresses and postal addresses. | Together with a name, they locate the person. |
| People and organisations | Names of people, company names and job titles. | A job title and a company are often enough to know who is meant. |
| Vehicles and devices | Number plates and IP addresses. | They are tied to a specific person or workstation. |
| Amounts and dates | A salary figure, the date of a period of sick leave or of a transaction. | Alone they say little; next to other data they point to someone. |
| Health data and other special categories | Diagnoses, medical reports, trade union membership. | Article 9 of the GDPR gives them reinforced protection. |
| Passwords and keys | Passwords, API keys, certificates. | They open systems; they should never leave their vault. |
| Confidential information | Contracts, bids, source code, internal minutes. | Even with no personal data, they are the company's or the client's secret. |
To see it on an example, the Leak detector in the LAB flags each piece of personal data in a text and returns a version with labels in their place. It runs on MagnolIA's servers, so try it with a sample text rather than a real one.
What does a one-page AI policy look like? A template to copy
Ten short clauses. The blanks in square brackets are what each company has to decide. Copy the text, fill them in and delete what does not apply to you.
Artificial intelligence acceptable use policy of [company name]
Purpose and scope. This policy applies to everyone who works at or for [company name], and to any artificial intelligence tool they use in their work, whether provided by the company or personal.
Approved tools. Only the tools on this list are used for work: [approved tools]. Anyone wishing to propose another writes to [contact person]; until it is approved, it is not used with company information.
Information that must not be entered. In tools that process data outside the company, nobody enters personal data of clients, employees or third parties (names, national ID numbers, phone numbers, email or postal addresses, bank accounts, card numbers, social security numbers, number plates), health data, passwords or keys, or [other confidential information: contracts, bids, source code].
Before pasting a text. If the task does not need to know who the document is about, it is anonymised first. If in doubt, do not paste it and ask [contact person].
Human review. What the AI produces is a draft. Whoever uses it reviews it, checks facts and citations, and is responsible for the result as for any other piece of their work.
Decisions about people. AI does not make decisions about people on its own: recruitment, appraisal, or granting or refusing a service. Any such use is first discussed with [data protection officer or legal adviser].
Transparency. When AI-generated content is sent to a client or published, [contact person] decides whether that has to be disclosed and how.
Accounts and access. Everyone uses their own account. Usernames and passwords are not shared, and email, calendars or other services are not connected to an AI tool without approval.
Training and incidents. The company explains this policy to every new joiner and repeats it [how often]. Anyone who believes they have entered information they should not have reports it to [contact person] as soon as possible.
Owner and review. [Role or person] is responsible for this policy. Approved by [management] on [date]. Next review: [date].
A starting model from MagnolIA, to be adapted with your data protection officer or legal adviser. It is not a legally valid document in itself and does not cover the obligations specific to each sector or country.
How do you adapt it and roll it out?
Find out what is already in use
Ask each team which AI tools it uses and what for. Without that inventory, the list in clause 2 will be fiction.
Decide the approved list
For each tool, find out where it processes data, how long it keeps it and whether it uses it for training. The questions to ask are in private AI for business.
Fill in the blanks with whoever handles data protection
Clauses 3, 6 and 7 above all: they depend on your data and your sector.
Explain it in person
With real everyday examples, and keeping a record that it was explained.
Give it a review date
Tools and rules change. A policy with no review date goes stale without anyone noticing.
If you work for the Spanish public sector, the policy sits alongside the awareness and training measures of the National Security Framework, explained in AI and Spain's ENS. And for law firms, it is worth reading together with AI for law firms.
What changes if the AI runs in your own network?
Clause 3 changes; the policy does not. With AI in the office, the document is processed on a machine in your network and is not sent to an outside provider, so the "never paste" list stops being a blanket ban for that tool, except for passwords and keys, which go into no tool at all. Giving the AI only the data the task needs still applies, and the other nine clauses are still needed: who has access, who reviews, what AI does not decide and whom to ask.
In a MagnolIA installation each person signs in with their own user, usage is logged per user, and the anonymiser replaces personal data when the task does not need it. That makes the policy easier to follow; it does not replace it.
Frequently asked questions about AI acceptable use policies
What should a company AI acceptable use policy include?
A company AI acceptable use policy should say, on one page, which tools are approved, what information must not go into them, who checks what the AI produces, whom to ask and when the text is reviewed. MagnolIA's template covers this in ten clauses, to be adapted with your data protection officer.
What data should you not paste into a public AI tool?
Avoid pasting into a public AI tool anything that identifies a person or is confidential to the business: names, ID numbers, phone numbers, email and postal addresses, bank and card details, social security numbers, number plates, health data, passwords or contracts. The Leak detector in MagnolIA's LAB flags the personal data in a text before you send it.
Is an AI acceptable use policy a legal requirement?
Neither the EU AI Act nor the GDPR mentions a document called an "AI acceptable use policy". The AI Act asks for measures supporting staff AI literacy (Article 4); the GDPR, for organisational measures appropriate to the risk (Article 32). A written policy helps to document them; whether it is enough is for your data protection officer to judge.
Who approves an AI acceptable use policy and how often is it reviewed?
An AI acceptable use policy should be approved by the company's management, after review by whoever is responsible for data protection and, if there is one, the security lead. It is sensible to review it at least once a year and whenever the approved tools or the rules change. MagnolIA's template leaves a blank for the next review date.
Sources
External figures on this page link to their original source.
EUR-Lex. Regulation (EU) 2024/1689, Artificial Intelligence Act: consolidated text of 27 July 2026. Version of 27 July 2026. eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:02024R1689-20260727
EUR-Lex. Regulation (EU) 2016/679, General Data Protection Regulation (GDPR). eur-lex.europa.eu/eli/reg/2016/679/oj
AEPD (Spain's data protection authority). The Agency publishes its internal policy on the use of artificial intelligence, the first of its kind in the public sector (in Spanish). 27 November 2025. www.aepd.es/prensa-y-comunicacion/notas-de-prensa/agencia-publica-politica-interna-uso-inteligencia-artificial
What does the policy look like with AI inside your network?
Tell us what documents you work with. We will show you what the policy looks like when the AI runs inside your network, and what would still be your job.
Request a pilot


