The short answer
An AI policy is the internal set of rules that says which artificial intelligence (AI) tools can be used in the company, with which data and with which checks. No law requires it under this name, but it is the simplest way to comply with the AI Act and the GDPR and to govern spontaneous use. For an SME one page is enough: permitted tools, data that must never be entered, checking of results, a person to ask.
- One clear page is worth more than twenty that nobody reads.
- Four essentials: permitted tools, prohibited data, checking of results, an internal point of contact.
- It is decided by the people who run the company, with those who know the processes, the systems and employment relationships.
- It works if the company offers safe tools: a ban with no alternative moves the use to where it cannot be seen.
- It must be reviewed when the tools, the providers or the law change.
Note. Informational content updated as of 12 October 2026. The national rules described are those of Italy; other EU member states have their own. The example rules are an outline to adapt, not a legal document: for privacy and employment matters, consult your advisers.
What it is, and why an SME needs one too
An AI policy, that is, a set of rules on the use of artificial intelligence (AI), answers a question that people in companies now ask every day: can I use this tool for this job?
Without a written answer, everyone decides for themselves. According to the Artificial Intelligence Observatory of the Politecnico di Milano, in Italy only 19% of those who use AI at work use company tools alone; among large companies, four in ten have set guidelines and 24% ban tools not supplied by the organisation. For Italian small and medium-sized enterprises there is no comparable figure: in an OECD survey of more than 5,000 SMEs in seven countries (Italy is not among them), only 28.6% of those that use generative AI have given their staff guidelines.
Rules are there so that you can say yes: people who know what is allowed use the tools with more peace of mind, and the company knows where its data goes.
Is it mandatory?
In general, no law requires a document with this name. There are, however, obligations that a policy helps to meet.
- AI Act. Article 4 requires those who use AI systems to take measures to support the AI literacy of their staff. Rules that are written down and explained are one of these measures. We cover this in the guide AI literacy: what the AI Act requires of companies.
- GDPR. If personal data ends up in a tool, the company must know who processes it, where and for how long.
- Employment. In Italy, to take one country’s example, Article 11 of Law No. 132/2025 (Italy’s AI law) obliges employers to inform workers when they use fully automated decision-making or monitoring systems that affect recruitment, management of the employment relationship, allocation of tasks, surveillance or evaluation of workers. Since 23 October 2026, Legislative Decree No. 179/2026 has added that final decisions on the employment relationship (hiring, changes, dismissal, disciplinary measures, evaluations) are for a person to take and cannot be taken by an automated system alone, otherwise they are void, and that AI systems which affect the organisation, pace or safety of work must be assessed in the workplace risk assessment document, with information and training for workers on the specific risks.
- Contracts. The confidentiality agreements signed with customers also apply to what an employee pastes into an assistant.
The guidelines of Italy’s Ministry of Labour on AI in the world of work, adopted in December 2025, are not binding but point in the same direction: they recommend that businesses have clear policies on the use of generative AI, accompanied by training, and they contain a handbook designed for SMEs and professionals.
What it must contain
Ten points. The first four are essential.
- The permitted tools. A list with names: which assistants, which features of the ERP (the company’s management software), with which type of account. Anything not on the list is not used for work until it has been approved.
- The data. What can go into the tools and what cannot. This is the point that matters most: it is covered further on.
- Checking the results. What AI produces is a draft. Whoever uses it is responsible for it: before using it, they check figures, names, dates, quotations and references to laws. Models can make up answers that look correct.
- The person to ask. A name, not a department.
- Who it applies to. Employees and external collaborators who work with the company’s tools.
- The uses that require authorisation. Connecting an agent to company systems, processing large amounts of data, using AI in decisions that concern people. Recruiting or evaluating workers with AI is a high-risk use: it is not done without a specific assessment.
- Transparency. When to tell colleagues and customers that a piece of content was produced with AI, in particular if it reworks material they supplied.
- How to ask for a new tool. Who to ask, and how soon the answer comes. If the official route is slow, people find another one.
- What to do if you make a mistake. Anyone who has entered confidential data by mistake says so straight away, with no consequences for the mere fact of having said so. A mistake that is reported can be dealt with; a hidden one cannot.
- Date, version and review. Who approved the document and when it will be reread.
The data: three colours
A simple classification works better than a long list.
| Colour | What it covers | Where it can be used |
|---|---|---|
| Green | Public or non-sensitive information: laws and regulations, catalogues, texts already published, market prices, drafts with no real data | In all the permitted tools |
| Amber | Internal information: procedures, quotes without confidential prices, ordinary correspondence, working documents | Only in company tools, with a contract that rules out training on the data |
| Red | Personal data of customers and employees, purchase prices, confidential price lists and margins, contracts, technical drawings and documents, material covered by confidentiality agreements, credentials | Only in systems expressly approved for that data, or never |
The handbook of Italy’s Ministry of Labour lists precisely these categories among the content not to be uploaded without adequate safeguards: personal data and information on employees, price lists, quotes and margin analyses, technical documents and designs, material protected by confidentiality agreements. When a piece of content is needed but sensitive, the way to go is to remove names, amounts and references from it before using it.
Where the data travels with each type of tool, and what the providers’ contracts promise, is explained in the guide AI in the cloud or on-premise.
A one-page example
This is an outline for a company of around fifty people. It must be adapted: the names of the tools and of the people are to be replaced.
Rules for the use of artificial intelligence
1. Why these rules. We want to use AI where it saves time, without putting at risk the data of the company, of its customers and of our colleagues.
2. Permitted tools. Only the tools on this list may be used for work, with the company account: [company assistant], [AI features of the ERP]. Personal accounts must not be used for work activities.
3. Data. Public information: in all the permitted tools. Internal information: only in company tools. The following must never be entered into any tool that has not been approved for the purpose: personal data of customers and employees, purchase prices, confidential price lists, contracts, technical drawings, documents received from customers under a duty of confidentiality, passwords.
4. Results. What AI produces is a draft. Whoever uses it checks figures, names, dates and references, and is responsible for it as for any other piece of their work.
5. Decisions about people. AI must not be used to recruit, evaluate or monitor workers; no decision on the employment relationship is taken by a system on its own (in Italy this is also the law).
6. New tools and automations. Anyone who wants to use a tool that is not on the list, or to connect AI to company systems, must ask [name]. The answer comes within five working days.
7. Mistakes. Anyone who enters confidential data by mistake must report it to [name] straight away.
8. Training. Everyone who uses these tools attends the basic session; those who join later catch up on it.
9. Who to ask. [Name and contact details].
Approved by senior management on [date]. Version 1. Next review: [date].
Who writes it, who approves it
- The people who run the company decide it. Establishing what can be done with company data is a choice for senior management, as we explain in the guide AI adoption in companies: top-down or bottom-up.
- A point of contact prepares it, with whoever looks after the IT systems and after consulting the heads of department: they are the ones who know which data circulates and which tools are already in use.
- The advisers look at the sensitive points: the privacy adviser for personal data, the employment law adviser for everything that touches on informing and monitoring workers.
- People are involved. The Italian Ministry of Labour’s guidelines suggest involving workers and their representatives in adoption processes. Rules that have been discussed are respected more than rules handed down from above.
A good test: have three people from different departments read the draft and ask them what is not clear.
How to get it followed
With three things, in this order.
- Tools that work. If the company tool is worse than the one people have on their phones, they will use their phones.
- A spoken explanation. A meeting in which the rules are read together, with examples from everyday work, is worth more than a signature to confirm they have been read.
- A point of contact who can be reached. Anyone with a doubt must be able to clear it up the same day.
Monitoring calls for caution. Recording and checking how employees use the tools touches on the rules on remote monitoring of workers: in Italy, to take one country’s example, those of Article 4 of the Italian Workers’ Statute. Tools that can result in monitoring may be used only for organisational and production needs, for workplace safety or to protect company assets and, unless they are the tools people work with, only after an agreement with the trade unions or, failing that, authorisation from the labour inspectorate (Ispettorato del lavoro). The privacy rules apply as well: the information collected may be used only if the worker has been adequately informed about the tools and the checks. Before switching them on, consult the employment law adviser.
The questions to ask before using a tool
The Italian Ministry of Labour’s handbook suggests some questions to ask yourself before uploading content to an AI service. In brief:
- Am I using a professional version or a personal account?
- Have I switched off the use of my data for training?
- Does the document contain personal, confidential or strategic data? Can I make it anonymous?
- Where is the data processed?
- Will the result need to be checked by an expert?
- Is the use consistent with professional obligations and with the rules of the sector?
The most common mistakes
- Copying a template found online without adapting it to the company’s tools and data.
- Writing twenty pages that nobody will read.
- Banning everything without offering an alternative.
- Listing the prohibitions and forgetting what can be done.
- Not naming a person to ask.
- Not updating it: a policy that mentions tools from two years earlier loses credibility.
How we work
A proposal for usage rules is one of the documents we deliver after the day with senior management for the process map and priorities, together with the map, the data inventory and the step-by-step plan. We write it by looking at the tools the company really uses and the data it handles. It is one of the results of our AI consulting for companies.
We apply the same rules to the systems we build: a limited set of tools, confirmations before writing, an action log. We describe them in the guide AI agents: what they are and what they can do.
Sources
- Italy's Ministry of Labour and Social Policies, Ministerial Decree No. 180 of 17 December 2025: guidelines for the implementation of AI in the world of work, with the handbook for SMEs and professionals (in Italian)
- Regulation (EU) 2024/1689 (AI Act), Article 4, as amended by Regulation (EU) 2026/1744
- European Commission, AI literacy: questions and answers (updated as of 27 July 2026)
- Regulation (EU) 2016/679 (GDPR)
- Italy's Law No. 132 of 23 September 2025, Article 11, on Normattiva (in Italian)
- Law No. 300 of 20 May 1970 (the Italian Workers' Statute), Article 4, on Normattiva (in Italian)
- Artificial Intelligence Observatory, Politecnico di Milano, press release of February 2026 (in Italian)
- OECD, Generative AI and the SME Workforce: New Survey Evidence (2025)
- Italy's Legislative Decree No. 179 of 7 October 2026, powers of the national AI authorities and penalties, Gazzetta Ufficiale No. 234 of 8 October 2026 (in Italian)
Checked on 28 October 2026. Rules and deadlines change: always check the official source before you decide.