2 August 2026

Mandatory AI training: what Article 4 of the EU AI Act really asks for (and what is noise)

You may have received an alarming email: "AI training is now mandatory!", "million-euro fines if you don't train your team!". As usual, there is a part that is true and a thick layer of fear marketing. Let us take it apart.

The part that is true

Article 4 of the European Artificial Intelligence Regulation (the EU AI Act) has applied since 2 February 2025, and it says something simple: every organisation using AI systems must take measures to support the development of AI literacy among the people who work there, proportionate to the tools it uses and the context of use.

And yes: that includes you. There is no minimum size — it applies equally to a multinational and to a sole trader — and "using AI" covers things as everyday as your team using ChatGPT to draft emails or quotes. If someone at your company uses AI to do their job, Article 4 applies to you.

What the rule does NOT say

This is where the commercial noise overshoots. The rule does not require any particular course: it sets no hours, no format, and demands no "accredited" certificates (no official accreditation exists for this — be wary of anyone selling you one). It asks for proportionate measures: what someone using a chat tool to write needs to know is not the same as what someone supervising a system that makes decisions about people needs to know.

And on those million-euro fines: the EU AI Act's heavy penalties are for prohibited practices and high-risk systems, not for Article 4. What does arrive in August 2026 are the surveillance authorities who can start asking questions — and the day you face an inspection, a data incident or a tender that asks about compliance, having it done and documented will make the difference.

So what needs doing? Three things

First: know what you use. Make the honest list: which AI tools the people in your team touch, what for, and with what data. That analysis is the basis of the "proportionality" the rule asks for.

Second: train accordingly. Not a generic "what is AI" course to tick a box, but training sized to what you actually use: where those tools get things wrong, which data must never go into them (your clients' data, for a start), how to spot the scams that arrive, and what the house rules are.

Third: write it down. This is the step almost everyone skips and the one that genuinely protects you: a record of who was trained and when, on what, and why those measures are appropriate in your case. Without paper, in the eyes of an inspection it is as if you had not done it.

The comparison that makes it easy to grasp

Think of workplace health and safety training: nobody expects an office worker to take the same course as someone working at height, and what counts is that the company can show each person received the training that fits them. AI literacy works the same way — and like health and safety, it is better to have it done before anyone asks for it.

I put together AlfabetIA, AI literacy material for small companies, with exactly this structure: proportionate training, a test, certificates and a compliance dossier ready to show. I mention it as an example of what Article 4 training should contain, not as an offer.