Article 4 of the EU AI Act has been in force since 2 February 2025. It requires companies that use AI to ensure their staff have a sufficient level of AI literacy. That applies not just to tech corporations, but to every company where someone uses ChatGPT for emails, where Copilot sits inside the Office suite, or where a vision system hangs on the production line.

I see two reactions to this in conversations. Some have never heard of Article 4. Others got an email from their legal department months ago and have been pushing the topic ahead of them ever since, because it is unclear what concretely needs to be done.

Both are understandable. So here is the short version from practice – with an important note up front: this is not legal advice. I am a technician, not a lawyer. The binding legal assessment of your AI use is a matter for your legal department or law firm. What I can contribute is the practical implementation of the training obligation.

What Article 4 actually requires

The text is short. In essence: providers and deployers of AI systems must take measures to ensure that their staff, and anyone working with AI systems on their behalf, have a sufficient level of AI literacy. Technical knowledge, experience, education and the context of use must be taken into account.

Three things stand out:

"Deployer" means: using is enough. You do not have to develop an AI system to fall under Article 4. Anyone who uses AI in their company is covered.
The required level depends on context. An administrator drafting texts with a chatbot needs different competence than a maintenance technician who is supposed to trust a predictive maintenance system. An off-the-shelf training that is identical for everyone misses the point of the text.
The how is open. The regulation prescribes no format, no number of hours and no certificate. What has become established in practice: documented training with proof of participation. Not because it is mandatory, but because otherwise you have nothing in hand when someone asks.

Does this apply to my company?

The most honest test question: does anyone at your company use AI? And not just officially.

In almost every company I know, the answer is yes – often without management being fully aware of it. The clerk has ChatGPT draft quotations. The junior in production planning translates data sheets with an AI tool. Microsoft has pushed Copilot into the programs your people work with every day anyway. This is called shadow AI, and it is the real reason the training obligation makes sense: the usage is happening regardless. The only question is whether it happens with or without rules.

On top of that come the official systems: quality inspection with computer vision, a RAG system over your maintenance manuals, predictive maintenance on the line. If you operate anything like that, you are clearly within scope.

What "sufficient AI literacy" means in practice

From my perspective – and this is how I implement it in trainings – a manufacturing company does not need a lecture on neural networks. The workforce needs to understand four things:

What AI can and cannot do. Realistic expectations instead of science fiction in either direction. A language model is not a knowledge store, a vision system is no guarantee of zero defects.
Where the risks lie in everyday work. Hallucinations: the model answers fluently and wrongly. GDPR and trade secrets: whatever you type into a public tool has left the company. Shadow AI: which tools are allowed, which are not.
Concrete rules for one's own role. What may administration do with customer data, what may QA do with inspection data, what applies on the line. Rules that fit on a poster get followed. Twenty-page policies do not.
When the human decides. AI outputs are suggestions. Whoever approves, checks. That sounds banal, but it is the core of what the regulation means by competence.

Who should be trained

The obvious answer would be: only those who officially work with AI. In practice, a mixed group works better – production management, plant management, QA, maintenance and administration together, in groups of up to about twelve people.

The reason is simple: AI use does not respect department boundaries. If everyone has heard the same rules, there are no discussions afterwards about what applies. And the leadership level needs to know what to expect from the tools it approves or bans.

The proof: without documentation, it didn't happen

A training that is not documented anywhere does not exist from a compliance point of view. What belongs in the file:

Who was trained and when (participation certificates)
What was covered (agenda or materials)
Who delivered the training and with what qualification

This is not bureaucracy for its own sake. If something goes wrong – data leaking through an AI tool, a wrong approval based on an AI answer – the first question is: were the people trained? At that point you want to be able to open a folder instead of improvising.

And if you do nothing?

The honest answer: Article 4 does not appear in the regulation's catalogue of fines as a separate offence. Anyone hoping for, or fearing, the quick fine bogeyman is wrong in both directions.

Still, sitting it out would be the wrong conclusion. The training obligation is applicable law, supervisory authorities and courts will refer to it in a dispute, and in the event of damage, the training that never happened is a remarkably bad argument. Add the practical point: the risks at stake – data leakage, hallucinations, uncontrolled tools – cost money even without any authority involved. The training pays off even if nobody ever asks about it. How this is to be assessed legally in your specific case is something to clarify with your law firm – see above, not legal advice.

How I implement this

I offer a half-day in-house training for this, about four hours, on site in Tyrol and western Austria or remote. Hands-on from the production floor instead of the lecture hall, for mixed groups of up to twelve people, with documented proof of training for your compliance records. Fixed price of €2,400 flat, including materials and participation certificates.

All details, contents and frequently asked questions: AI training under Art. 4 EU AI Act

Or request a date directly – I will get back to you with an honest assessment of whether and how this fits your company.

Related:

EU AI Act Article 4: What Manufacturing Companies Must Train For Now