The short answer
Yes, artificial intelligence (AI) training is an obligation, but not a course with a certificate. Article 4 of the AI Act, applied since 2 February 2025 and rewritten in July 2026, requires those who provide or use AI systems to take measures to support staff AI literacy, with no set hours, programmes or levels; it applies even if you only use ChatGPT. It is a decision for the people who run the company, written into the plan and ready to show.
- The obligation has existed since 2 February 2025 and concerns anyone who uses AI systems at work, even just an assistant.
- Since 27 July 2026 a "sufficient" level is no longer required: what is required is measures to support AI literacy.
- No prescribed course and no certificate: an internal record of what has been done is the way the Commission indicates for documenting it.
- The measures must be proportionate to roles, tools and risks.
- It is a decision for the people who run the company: it belongs in the adoption plan, alongside objectives, priorities and usage rules.
- In Italy, as one country's example, supervision lies with the National Cybersecurity Agency (ACN), with powers and fines set by Legislative Decree No. 179/2026, in force since 23 October 2026: no fine is provided for AI literacy.
Note. Informational content updated as of 12 October 2026, based on the official texts cited below. The national rules described are those of Italy; other EU member states have their own. This is not legal advice: for compliance decisions, consult a professional.
The short answer
Yes, there is an obligation. No, it is not an obligation to buy a course.
It is a decision for the people who run the company, not a compliance task to be delegated to a course: the AI adoption plan, alongside objectives, priorities and usage rules, must say who needs to know what, about which tools, and how the company documents it.
Article 4 of the AI Act, Regulation (EU) 2024/1689, requires providers and deployers of artificial intelligence (AI) systems to take measures to support the AI literacy of their staff. AI literacy is the term the Regulation itself uses. It has applied since 2 February 2025.
In July 2026 the article was rewritten. Many pages found online still describe the previous, stricter version, and talk about levels to be ensured and automatic fines. This guide follows the text in force today and the clarifications published by the European Commission.
What Article 4 says today
The text in force since 27 July 2026 says, in substance, three things.
- Providers and deployers take measures to support the development of AI literacy among their own staff and the other people who use the systems on their behalf. In doing so they take into account those people’s technical knowledge, experience and training, and the context in which the systems are used.
- No specific level of AI literacy has to be guaranteed for any individual. It is written in the text.
- The Commission and the member states help businesses, SMEs in particular, including by publishing practical examples of how to comply.
| Aspect | Until 26 July 2026 | Since 27 July 2026 |
|---|---|---|
| What is required | To ensure, to their best extent, a sufficient level | To take measures to support the development of AI literacy |
| Level to be reached | “Sufficient” | No specific level |
| Public support | Not provided for in the article | Practical examples and European recommendations, with attention to SMEs |
| Is it an obligation? | Yes | Yes |
So the obligation remains. What has changed is how it is measured: no longer a result to be ensured, but an effort to be demonstrated.
Who it applies to
To almost every company that uses artificial intelligence tools.
Anyone who uses an AI system under their own authority is a deployer: that is the Regulation’s term. You do not need to have developed anything: it is enough to use an assistant, a translation program, an AI feature inside the ERP (the company’s management software). The European Commission gives precisely the example of a company whose employees use ChatGPT to write or translate texts: the obligation applies, and people must be informed about the specific risks, starting with made-up answers.
The measures concern staff and the other people who use the systems on the company’s behalf: an external collaborator, a contractor, a service provider, a client to whom the company makes the system available.
What “measures” means
The Regulation does not list them. The Commission indicates four steps to consider.
- A general understanding of AI in the company. What it is, how it works, which tools are used, what opportunities and what dangers they bring.
- The company’s role. Does it develop AI systems, or use systems developed by others?
- The risk of the systems used. What the people who operate them need to know, and which risks they must be able to recognise and reduce.
- Actions built on this analysis, taking into account the differences between people and the context: in which sector, for what purpose, on whom the effects fall.
Two practical consequences follow. Training can be different for different roles: the Commission explicitly allows this. And having people read the instructions for use is not enough: in many cases, the Commission writes, it may be ineffective.
Do you need a course? Do you need a certificate?
A course, not necessarily. There are no mandatory formats and no imposed programmes. What counts is that people receive training and guidance suited to their role and to the tools they use.
A certificate, no. No certificate is needed. Organisations can keep an internal record of training activities and other initiatives.
An AI officer, no again. Article 4 does not impose a governance structure or a dedicated role.
Measuring what each person knows, not that either. The article does not oblige you to check what each employee knows.
Who supervises, and what you risk
Supervision lies with the national market surveillance authorities, not with the AI Office, and started at the beginning of August 2026. The penalties for this obligation are based on national laws and must be proportionate: what counts is the nature and gravity of the infringement and whether it is intentional or due to negligence. The Commission notes that a penalty is more likely when an incident turns out to be linked to a lack of adequate training and guidance.
In Italy, to take one country’s example, Law No. 132/2025 (Italy’s AI law) designates the National Cybersecurity Agency (ACN) as the market surveillance authority and the Agency for Digital Italy (AgID) as the notifying authority. Italy’s Legislative Decree No. 179 of 7 October 2026, published in the Gazzetta Ufficiale (the official gazette) on 8 October and in force since 23 October 2026, sets out their powers and the penalties. It provides for no fines for AI literacy: the financial penalties concern prohibited AI practices, the obligations of providers and deployers of high-risk AI systems, transparency and incorrect information to the authorities. For minor infringements the authority can order a remedy within a set time; anyone who does not comply pays the fine, increased by up to one third. The decree does, however, ask employers for two things that concern training: to assess, in the workplace risk assessment document, the AI systems that affect the organisation, pace and safety of work, and to give workers information and training on the specific risks of these systems (Article 42). It asks Italy’s professional bodies (the ordini professionali) to include AI literacy in their mandatory training, with a minimum number of hours (Article 47).
Then there is liability towards anyone who suffers damage. The AI Act does not create a right to compensation, but if an error arises from the unprepared use of a tool, the lack of training can weigh in court proceedings under national rules. In Italy, since 30 September 2026, Legislative Decree No. 160/2026 has provided that when the damage results from a breach of obligations under the AI Act, the causal link between the breach and the damage is presumed, unless proven otherwise: we cover this in the guide on the AI Act.
The other obligations, not to be confused with this one
- Human oversight of high-risk AI systems. Anyone who uses high-risk AI systems, for example to recruit or evaluate workers, must assign oversight to competent, trained people. These obligations apply from 2 December 2027.
- Information to workers. In Italy, Article 11 of Law No. 132/2025 obliges the employer to inform workers when it uses fully automated decision-making or monitoring systems that affect recruitment, management of the employment relationship, allocation of tasks, surveillance or evaluation of workers. It is an obligation to inform, different from the obligation to train.
- Decisions on the employment relationship. In Italy, since 23 October 2026 (Legislative Decree No. 179/2026, Article 41), final decisions on hiring, on changing or ending the employment relationship, disciplinary measures and evaluations cannot be taken by an automated system alone: a person decides, and a dismissal in breach of this rule is void.
- Training on health and safety risks. Article 42 of the same Italian decree requires the employer to give workers information and training on the specific risks of AI systems that affect the organisation, pace and safety of work. It is an obligation of the workplace safety system, different from AI literacy under the AI Act, and it is documented in the same register.
The full picture is in the guide AI Act: what a manufacturing company must do.
What to do, in practice
For those who run a company that uses assistants and a few AI tools in its offices, five decisions are an adequate starting point. They are the same ones that go into the adoption plan.
- List the tools. Those bought by the company and those that people use on their own initiative. It is the same survey needed to decide how to organise adoption.
- Say who uses what. Roles, tasks, data involved.
- Decide what the people who use the tools need to know. How they work, what they get wrong, which data must not be entered, how to check an answer before using it. Through an internal session, the provider’s material or an external course, depending on the roles: the company chooses.
- Go deeper where the risk is greater. Those who use tools that touch customers, workers or important decisions need more. If AI enters the production departments or changes the pace of work, training on the specific risks goes into the safety plan (in Italy, Article 42 of Legislative Decree No. 179/2026).
- Write a few rules and keep a record. One page of usage rules, as in the example in the guide Company AI policy, and a register of what has been done.
The measures are not decided once and for all: they need to be revisited when the tools change or new people arrive.
The guidelines of Italy’s Ministry of Labour on AI in the world of work, adopted in December 2025 and non-binding, point in the same direction: they tell businesses to draw up an inventory of the AI systems in use, name a contact person, and train and inform workers. For SMEs they also contain a handbook on the use of generative AI. Italy’s Legislative Decree No. 179/2026 does not make this role mandatory.
The register: what to record
There is no official template. A table like this one can be enough to document what the company has done.
| Date | Initiative | For whom | Content | Documents kept |
|---|---|---|---|---|
| 14 October | Basic session, two hours | Sales and administration (accounts) offices | How an assistant works, typical errors, data not to enter | Presentation, attendance list |
| 21 October | Usage rules handed to everyone | All staff | Permitted tools, prohibited data, who to ask | Text of the rules, confirmation of reading |
| 11 November | In-depth session | Those who use the agent in the ERP | Confirmations, checks, cases in which to stop | Operating instructions |
The dates and initiatives are an example.
Where companies stand in Italy
According to the Artificial Intelligence Observatory of the Politecnico di Milano, in Italy more than one large company in two, among those analysed, has started AI literacy initiatives, but only 15% have a structured project to comply with the AI Act. Among SMEs, according to the Digital Innovation in SMEs Observatory of the same Politecnico, only 7% have started structured AI training programmes.
False alarms
- “Without a certificate you risk a fine.” The Regulation does not ask for certificates.
- “You need so many hours per employee.” No number of hours is prescribed.
- “It only concerns those who develop artificial intelligence.” It also concerns those who use it.
- “Since July 2026 the obligation has been abolished.” It has been lightened, not abolished.
- “Getting people to sign a sheet of paper is enough.” What is needed is that people really know what to do and what to avoid.
How we work
We do not run courses. In our method, AI literacy of staff is one of the decisions senior management takes during the process map and priorities day, together with objectives, the order of the processes, usage rules and spending. After that day we deliver a proposal for usage rules, which is the first measure to record in the register, and in the step-by-step plan we write who needs to know what for each tool that enters the company. How to do it, through an internal session, the provider’s material or an external course, is for the company to decide. The path is described on the page AI strategy and consulting.
Sources
- Regulation (EU) 2024/1689 (AI Act), Articles 3, 4, 26 and 99
- Regulation (EU) 2026/1744 of 8 July 2026 (the Digital Omnibus on AI), which rewrote Article 4
- European Commission, AI literacy: questions and answers (updated as of 27 July 2026)
- European Commission, repository of AI literacy practices
- Italy's Law No. 132 of 23 September 2025, Articles 11 and 20, on Normattiva (in Italian)
- 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 (in Italian)
- Artificial Intelligence Observatory, Politecnico di Milano, press release of February 2026 (in Italian)
- Digital Innovation in SMEs Observatory, Politecnico di Milano, press release of 21 May 2026 (in Italian)
- 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)
- Italy's Legislative Decree No. 179 of 7 October 2026, text on Normattiva (in Italian)
Checked on 28 October 2026. Rules and deadlines change: always check the official source before you decide.