AI Act Article 4: What Is AI Literacy, and Does Your Company Need AI Training?

What does Article 4 require, in plain language?

Article 4 of the European AI Regulation (AI Act) provides that providers and users of AI systems must take measures so that their staff and other people acting on their behalf have sufficient AI literacy. Literacy here means something entirely practical: a person understands what the tool does, where it makes mistakes, what may be put into it, and how to check what comes out of it.

The article contains an important additional phrase: "taking into account their technical knowledge, experience, education and training, and the context in which the AI systems will be used." The requirement is not the same for everyone: a bookkeeper using AI to draft letters and an analyst processing client data need a different level of preparation.

Just as important is what the article does not require: there is no mandatory program, no state exam, no certification requirement and no set number of hours.

Does this apply to my company?

Most likely yes, and this is the most commonly misunderstood part of the requirement. The article applies both to providers (those who develop AI systems or place them on the market) and to users — any organization that uses AI in its operations, regardless of size or industry.

In practice: if a sales manager writes proposals with ChatGPT, or someone in logistics has AI organize a shipping list, you are a user of an AI system, even if nobody at the company calls it that. The requirement also covers "other persons acting on the company's behalf" — outsourced service providers and contractors.

There is also a non-legal risk hiding here: some employees already use AI privately and bring it to work with no rules at all, so-called shadow AI. The company is responsible for that use whether or not it has noticed it.

From what date does it apply?

The obligation is not new: Article 4 has applied since 2 February 2025. Anyone waiting for it "to start" is already too late — it has already happened.

What changes now: on 2 August 2026, the national market surveillance authorities gain formal supervisory and enforcement powers. In Latvia, VARAM is generally responsible for implementing the AI Act, but market surveillance is split between sector authorities, including the Consumer Rights Protection Center and the Data State Inspectorate (Datu valsts inspekcija).

Did the Digital Omnibus change anything?

It changed something, but not what many hoped. The Digital Omnibus Regulation (EU) 2026/1744 entered into force on 27 July 2026 and softened the wording of Article 4: the requirement to "ensure a sufficient level of AI literacy" became "take measures to support the development of AI literacy." That is a shift from an obligation of result to an obligation of effort — a specific level no longer has to be guaranteed, but proportionate action has to be shown.

Two things must not be misread. First, the article was not repealed: a proposal to turn it into a non-binding recital was rejected. Second, the date was not moved: the Omnibus postponed the obligations for high-risk systems to 2027 and 2028, but left Article 4 untouched.

What counts as compliance?

The European Commission's explanation states that there is no one-size-fits-all template, and no strict requirements or mandatory training are imposed. The assessment is risk-based: how risky your AI systems are, and what the human role and context are. In practice, companies that have this in order are able to show four things:

• A list of AI systems — which tools are used in the company, and in which departments.
• Role-appropriate training — different roles get preparation of different depth, instead of everyone getting the same presentation.
• Internal usage rules — what may be entered, how the output is checked, who to ask with questions.
• A training record — who learned what, and when. The Commission states this directly: organizations can keep an internal record of training.

Do you need a certificate?

No. The European Commission's explanation of AI literacy answers directly: a certificate is not necessary, and organizations can keep an internal record of training. Neither mandatory course content nor an exam is set.

This matters right now, because around the August date, offers are appearing whose main selling point is a "compliance certificate." A piece of paper like that does not, by itself, prove anything.

What should a company with 20-50 employees do — step by step

This is not a months-long project. Five actions are enough:

• Write down who uses what. In one table: department, tool, for which tasks. For most managers, this step is a surprise.
• Split people into roles by risk. Who uses AI for internal text, who for client communication, who touches personal data. Three groups are usually enough.
• Write a one-page set of rules. What must not be entered (client data, contracts, passwords, non-public financial information), how the output is checked, who to ask with questions.
• Run the training by role and keep the program, the date and the list of participants.
• Put a repeat in the calendar. A one-off session in 2026 will not be proof in 2028.

What should you not do?

• Do not panic. Article 4 does not have its own penalty category; penalties are set by member states and must be proportionate. The figure of "€35 million" applies to prohibited AI practices, not a lack of training — anyone who uses it in a conversation about training is scaring you, not explaining.
• Do not buy a "certificate" from an unknown source. A certificate is not required; an unverifiable certificate is not even proof.
• Do not forget contractors. If an outsourced provider uses AI on your behalf, they are in the same circle.
• Do not train only management. The risk is where the tool is used every day, not where it is decided on.
• Do not confuse a ban with a solution. A formal AI ban with no alternative does not stop the use, it just makes it invisible.

Where to check this yourself

This is not legal advice. For your specific situation, especially in regulated industries, it is worth discussing the question with a lawyer. Primary sources you can open: Article 4 of the AI Act and the European Commission's explanation of AI literacy.

Frequently asked questions