Company AI policy. A Word template for 10 to 50 people, covering the training and monitoring rules the law already asks for.
No law asks for a document by this name, yet the AI Act, Italian Law 132/2025 and the Workers' Statute ask for things a written policy holds together. Here is the template in Word, article by article.
No rule requires a company policy on artificial intelligence by this name: its job is to meet, and be able to prove, obligations that already exist. It says which tools are used, which data stays out and who approves a text before it leaves.
Article 4 of the AI Act has required AI literacy measures since 2 February 2025, in the wording rewritten by Regulation (EU) 2026/1744, and the European Commission says an internal record is enough. In Italy, Law 132/2025 requires employers to inform workers in the cases set by Article 1-bis of Legislative Decree 152/1997, meaning fully automated systems.
The template downloads below, in Word and PDF, with no email required: twelve articles, each with the rule it comes from. The file is in Italian.
Written by: Mattia Esposito, AI and marketing consultant, founder of Itria AI. From Puglia, based in Turin since 2023, he works remotely in Italy and abroad. LinkedIn profile.
The glossary entry on shadow AI offers a one-page set of rules to sign. This guide goes further: a full policy for a company of 10 to 50 people, with a tool inventory, worker information and monitoring rules. It is written for people who run a business, and it stops where an employment consultant's work begins.
The AI acceptable use policy template, to download
Twelve articles over three pages, with the choices to make in grey and the legal basis under each article, in Word to fill in and PDF to print. It suits a company using generative assistants and a few automations, with no in-house legal team. A note at the top says plainly that it is an information template, to be adapted.

| File | Contents | Link |
|---|---|---|
| Wordto fill in, in Italian | The twelve articles with the choices in grey, plus the tables for the inventory, the training log and the signatures. | |
| PDFto print, in Italian | The same policy, to read through in a meeting or fill in by hand. |
| Article | What you decide | Where it comes from |
|---|---|---|
| 1. Purpose | Who it applies to, who owns the rules, when they are reviewed. | AI Act, Art. 4 |
| 2. Inventory | Tool, provider, account, tasks, users, data processing agreement, where the data is held. | AI Act, Art. 4; GDPR, Art. 28 |
| 3. Tools | Company accounts only; systems that infer people's emotions are banned. | AI Act, Art. 5 |
| 4. Data | The traffic light: what can be used, what only with a data agreement, what never. | GDPR, Arts. 9 and 28 |
| 5. Sign-off | Who reads and approves outgoing texts, and who checks the figures. | The company's choice |
| 6. Customers | When to say there is a machine on the other side. | AI Act, art. 50 |
| 7. Training | The session on the tools and the log, with no certificates. | AI Act, Art. 4 |
| 8. Worker information | The yes-or-no question on automated systems, and what to do if the answer is yes. | Law 132/2025, Art. 11; Legislative Decree 152/1997, Art. 1-bis |
| 9. Monitoring | Who can see the prompt history, for what purpose, for how long. | Workers' Statute, Art. 4 |
| 10. Impact | When to ask for an impact assessment before starting. | GDPR, Art. 35 |
| 11. Incidents | Who to report to, and the 72 hours for notifying the data protection authority. | GDPR, Art. 33 |
| 12. Acknowledgement | The link to the disciplinary code, posting the rules, the signatures. | Workers' Statute, Art. 7 |
Itria's own rules on artificial intelligence are public, on two pages. On the ethics page, the third principle is headed “The machine prepares, but a person always decides”: anything that commits the company or touches someone's money waits for a person, message by message, even when that slows things down.
The AI transparency page does the job of Article 2 of the template. It lists every provider with what it processes and under which contract: Anthropic's model under a data processing agreement that rules out training on the data, and the orchestration running on servers in the European Union. The map of AI systems is available on request, and the Article 4 AI literacy note is listed as available.
Is a company AI policy mandatory?
Not by that name, but the obligations it contains are in force, some of them for years. They are spread across four texts: the AI Act, the GDPR, Italian Law 132/2025 with the decree it refers to, and the Workers' Statute. The table lines them up for a company that uses AI without building it.
| Obligation | For | Rule and date |
|---|---|---|
| AI literacy measures | Anyone using AI systems for work, even just ChatGPT, according to the European Commission. | AI Act, Art. 4; since 2 February 2025 |
| No systems that infer emotions at work | Every employer, except for medical or safety reasons. | AI Act, Art. 5(1)(f); since 2 February 2025 |
| Telling people they are talking to a machine | Anyone deploying a system that interacts with people, a website chat for instance. | AI Act, Art. 50; since 2 August 2026 |
| Informing workers and their representatives | An employer using high-risk systems, in recruitment for instance. | AI Act, Art. 26(7); with the high-risk obligations, postponed to 2 December 2027 |
| A contract with the data processor | Any company passing personal data to an AI provider. | GDPR, Art. 28 |
| Impact assessment | Processing with new technologies likely to result in a high risk. | GDPR, Art. 35 |
| Information on automated systems (Italy) | An employer using fully automated systems for hiring, duties or assessment. | Law 132/2025, Art. 11, since 10 October 2025; Legislative Decree 152/1997, Art. 1-bis |
| Agreement for monitoring tools (Italy) | An employer installing tools that make remote monitoring of work possible. | Workers' Statute, Art. 4 |
| Disciplinary code on display (Italy) | An employer wanting to enforce the rules with disciplinary measures. | Workers' Statute, Art. 7 |
For a small business using generative assistants to write, translate and summarise, the rows that really matter are the first, the fifth and the eighth. The others only apply with certain tools, which is why the template starts from the inventory: it shows which rows concern the company.
What an AI acceptable use policy for employees should contain
It has to answer six questions: who is covered, with which tools, which data, who signs off, what people know about monitoring, how they learn. Tools and data fall under the GDPR, monitoring under the Workers' Statute, training under Article 4 of the AI Act; sign-off is the company's call.
In its AI literacy answers, updated on 20 October 2026, the European Commission sets out a minimum path in four steps: understand what AI is and which systems the company uses, clarify your role (provider or deployer, the organisation using the system), assess the risk of those systems, and build actions on that analysis. The policy is where the four steps get written down.
Article 4 covers anyone using the systems on the company's behalf, so contractors and interns too: Article 1 of the template names them all, and sets a review every six months.
Approved tools: company accounts only
For work, staff use only the tools in the inventory, with a company account. The reason is Article 28 of the GDPR: when a provider processes personal data on the company's behalf, the company may use only processors that provide sufficient guarantees, under a contract. A free personal account has no such contract, and breaches of the obligations in Articles 25 to 39 of the GDPR, Article 28 among them, carry fines of up to 10 million euros or 2% of worldwide annual turnover, whichever is higher (Article 83(4)).
Read the data terms in the business plan's contract before signing: whether the provider trains models on your content, where it stores it, who can access it. How the plans of the main assistants differ, and where the data ends up, is covered on the page about ChatGPT, the GDPR and the Cloud Act and in the comparison of ChatGPT, Claude and Gemini for work.
The template adds a ban many policies leave out. Since 2 February 2025 Article 5 of the AI Act has prohibited using AI systems to infer the emotions of a person in the workplace, except for medical or safety reasons. A plug-in that gauges colleagues' mood from their chats is out, even when it is sold as a wellbeing tool.
What data never goes into AI: the traffic light
The simplest way to write the data rule is to sort data into three colours, each with the tools it may go into. When in doubt, the data is red and you ask the policy owner. The table is Article 4 of the template.
| Level | What data | Where it may go |
|---|---|---|
| Green | The company's public texts, drafts without names, product sheets already published, anonymised or made-up data for a test. | Any tool in the inventory |
| Amber | Names and contact details of customers and suppliers, orders, quotes, internal documents that are not confidential. | Only tools with a data processing contract |
| Red | Health, opinions, union membership and the other special categories in Article 9 of the GDPR; colleagues' data; passwords and keys; confidential documents. | In no tool at all |
Colleagues' data sits in the red for a specific reason: an employee pasting an appraisal or a sick note into an assistant takes another person's data outside the purpose for which the company holds it. The rest of the rule is practical. Remove the name before asking, use made-up examples, work on the public text.
Who signs off a text before it goes out
Any text written with AI that goes to customers, suppliers or public bodies is read and approved by a person before it is sent, and the sender answers for it as if they had written it themselves. Generative assistants can state false facts with confidence, the so-called hallucinations: figures, dates and legal references are checked against the source.
The template asks for two names: who approves offers and quotes, and who approves communications to public bodies and customers. This is human oversight decided in advance. Towards customers, Article 50 of the AI Act applies from 2 August 2026: people talking to an automated chat must be told, and realistic generated content must be disclosed. The detail is on the page about the AI Act and generated content.
AI literacy under Article 4: what goes in the training log
Article 4 of the AI Act has applied since 2 February 2025, and since July 2026 it asks companies to take measures supporting the skills of people who use AI, with no level fixed by law. It carries no penalty of its own: the band of up to 15 million euros or 3% of turnover in Article 99(4) lists Articles 26 and 50, and Article 4 is not among them.
Regulation (EU) 2026/1744, published on 24 July 2026 and in force from the third day after, rewrote it. The AI Act Service Desk now gives the new wording: “Providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff”. The new text adds that no specific level of AI literacy has to be ensured for any person.
The proof stays in-house. The European Commission puts it plainly: “There is no need for a certificate. Organisations can keep an internal record of trainings and/or other guiding initiatives.” No certificate, and no AI officer to appoint. What you do need is to explain the specific risks of the tools: the Commission gives the example of staff using ChatGPT for advertising copy or translations, who should be told about hallucinations.
The log in Article 7 of the template has five columns: date, tools covered, risks explained, attendees, who ran the session. For six people one session and one line will do; for fifty, separate groups by role. How deep the training should go, and who supervises it in Italy, is in the guide to mandatory AI Act training.
Informing workers under Italian Law 132/2025
Italian Law 132/2025 requires employers to inform workers about their use of AI, but only in the cases set by Article 1-bis of Legislative Decree 152/1997. Article 11(2) says the employer or principal “is required to inform the worker of the use of artificial intelligence in the cases and in the manner” of that decree (our translation).
Those cases are narrow. Since 5 May 2023 Article 1-bis has covered “fully automated decision-making or monitoring systems” (our translation) that give indications on hiring, managing or ending the relationship, assigning tasks, surveillance, assessment and performance. On a plain reading, an assistant used to draft an email decides nothing on its own and does not appear to fall within it; software that assigns shifts or rates performance without human input does.
If the company uses one of those systems, the notice must be given before the work starts and covers six points: the aspects of the relationship the system affects, its purposes, its logic and functioning, the categories of data and main parameters, the control measures and who is responsible for quality, and accuracy, robustness, cybersecurity and any discriminatory effects.
The same notice goes to the workplace union representatives; any change is communicated in writing at least 24 hours in advance; workers' requests get a written reply within 30 days.
Article 8 of the template turns all of this into a yes-or-no question, asked again at every review. The other parts of the Italian law, from the professions to copyright, are in the guide to Law 132/2025.
Remote monitoring: Article 4 of the Italian Workers' Statute
An AI assistant given to an employee for work is, as a rule, a tool for carrying out the job, and needs no union agreement. Article 4(2) of the Workers' Statute: “The provision in paragraph 1 does not apply to the tools used by the worker to carry out the work” (our translation).
Two points remain. Assistants keep a history of prompts, and paragraph 3 allows that data to be used in the employment relationship only if the worker has received adequate information on how the tools are used and monitored: Article 9 of the template is there for this. And a tool that also serves to monitor what people do falls under paragraph 1, with a union agreement or, failing that, authorisation from the National Labour Inspectorate.
The Italian Data Protection Authority has already drawn the line. On 14 May 2026 it warned a company selling a plug-in for Slack and Teams that gauged employees' emotions and stress from their messages: such systems, it wrote, must not lead to “making available to employers information, inferred through artificial intelligence systems, about their staff” (our translation).
This was a warning rather than a fine. In the same decision the Authority recalls the AI Act ban on inferring emotions at work. How the two sets of rules intersect is on the page about the AI Act and the GDPR.
Italy, the European Union, the United Kingdom and the United States
AI literacy and the bans come from the AI Act and apply across the European Union; worker information and monitoring are Italian rules. A company with a site or a partner abroad keeps one policy per country, or a shared one with the national parts kept separate.
| Where | What applies to AI at work | Texts |
|---|---|---|
| Italy | The EU rules, plus information on fully automated systems and an agreement or authorisation for monitoring tools. | Law 132/2025; Legislative Decree 152/1997; Workers' Statute |
| European Union | AI literacy measures, the ban on inferring emotions at work, information before high-risk systems, a contract with data processors. | AI Act; GDPR |
| United Kingdom | No AI Act. UK data protection rules still apply, including to worker monitoring. | UK GDPR |
| United States | No federal requirement to have an AI policy; state and local rules, mostly on recruitment tools. | State and city rules |
For a group or a foreign partner, the shadow AI entry also offers a one-page English set of rules, an AI acceptable use policy. The twelve-article template is built for Italian law.
How to adopt it in a company of 10 to 50 people
In six steps, and the first is asking who uses what, before writing a single line. A policy written without an inventory bans tools nobody uses and misses the ones actually in use.
The round of questions can be done face to face or with an anonymous form, with no consequences for anyone. Then pick one or two tools on a business plan, fill in the template, post it alongside the disciplinary code and give everyone a copy. The training session goes into the log the same day. After six months, or with the first new tool, start again from the inventory.
If the company already runs automations, the most useful step is to draw where they act before deciding who approves them: the process map is for exactly that. To choose the tool at the centre well, see the guide to ChatGPT projects; the other guides are collected in the index for small businesses.
Company AI policy or internal regulation?
They are the same thing under two names: policy is the term used by international groups, regolamento the one most Italian companies use. In English people search for an AI acceptable use policy or a company AI policy.
The difference that matters lies elsewhere. A policy published on a website speaks to customers and suppliers, like Itria's ethics and transparency pages; an internal policy speaks to employees, and to enforce it in the employment relationship in Italy it must be posted where they can see it, as Article 7 of the Workers' Statute requires. Many companies need both.
Questions and answers
Is a company AI policy mandatory in Italy?
Not by that name: no Italian or EU rule requires one. The rules do ask for things a policy holds together: AI literacy measures for anyone using AI (Article 4 of the AI Act, since 2 February 2025), a contract with any provider processing data on the company's behalf (Article 28 GDPR), and information on monitoring under Article 4 of the Workers' Statute.
Putting them in one document is the simplest way to comply and to prove it.
Do you have to tell employees if the company uses ChatGPT?
The information duty in Law 132/2025 points to Article 1-bis of Legislative Decree 152/1997, which covers fully automated decision-making or monitoring systems affecting hiring, duties, surveillance and assessment. An assistant used to draft an email does not, on its own, appear to fall within it.
Two things remain: AI literacy under Article 4 of the AI Act, which the European Commission explicitly extends to ChatGPT users, and information about conversation logs if the company intends to use them.
Do you need a union agreement to give staff an AI assistant?
As a rule, no, if the assistant is a tool the employee uses to do the job: Article 4 of the Workers' Statute exempts such tools from the agreement. A union agreement, or authorisation from the National Labour Inspectorate, is needed for a tool that also serves to monitor what people do.
Either way, the data collected can be used in the employment relationship only if the worker has received adequate information.
Do employees need a certified AI course?
No. The European Commission says no certificate is needed and an internal record of training and other initiatives is enough. Nor do you need to appoint an AI officer.
Since July 2026 Article 4 asks companies to take measures to support skills, without guaranteeing a specific level for each person. The template's training log has five columns: date, tools, risks explained, attendees, who ran the session.
What happens if an employee breaks the policy?
The consequences are those in the company's disciplinary code and the applicable collective agreement. For them to apply, Article 7 of the Workers' Statute requires disciplinary rules to be brought to workers' attention by posting them in a place everyone can reach.
The template covers this in its last article, together with a signed acknowledgement. For a specific case, ask an employment consultant.
Notes on sources
- Regulation (EU) 2024/1689, AI Act: Arts. 4, 5(1)(f), 26(7), 50 and 99(4). Regulation (EU) 2026/1744, the Digital Omnibus on AI, Official Journal of 24 July 2026: new text of Art. 4, source of the quotation. Texts read on 13 November 2026 from the EU Publications Office portal.
- European Commission, AI Act Service Desk, Article 4: AI literacy, and AI Literacy, Questions & Answers, updated on 20 October 2026: the four steps, the internal record, the ChatGPT case. Read on 13 November 2026.
- Regulation (EU) 2016/679, GDPR: Arts. 9, 28, 33, 35 and 83(4). Read on 13 November 2026.
- Italian Law no. 132 of 23 September 2025, Official Gazette no. 223 of 25 September 2025, Arts. 11 and 12; in force since 10 October 2025. Read on 13 November 2026 in the text published in the Gazette.
- Italian Legislative Decree no. 152 of 26 May 1997, Art. 1-bis, text in force since 5 May 2023. Law no. 300 of 20 May 1970, the Workers' Statute, Art. 4 and Art. 7. Read on Normattiva on 13 November 2026.
- Italian Data Protection Authority (Garante per la protezione dei dati personali), decision of 14 May 2026, doc. web no. 10255494: a warning, source of the quotation. Read on 13 November 2026.
- The Itria rules cited are public on the Ethics and AI transparency pages, read on 13 November 2026.
A policy says who decides. A well-built system asks them, before every send. The system first, then the growth.
Mattia Esposito, who wrote this guide, builds tailored AI systems for small businesses with Itria: automations connected to the software the business already uses, and a Dashboard showing what has happened and what is waiting for a decision. Nothing goes to a customer without a person’s yes. On a test bench on 4 September 2026, with 30 documents including orders, delivery notes and invoices, the system extracted 234 fields out of 237 correctly, and the unreadable document deliberately placed in the batch was set aside without a single invented field.