AI literacy: since July, article 4 asks something different of you

If you work with AI, you also need to make sure your people know how to handle it. That's been in Article 4 of the AI Act since February 2025, the article on AI literacy. This summer, that article was amended, and the rule got simpler.

What changed

This was the old text:

"Providers and deployers of AI systems shall take measures to ensure, to their best extent, a sufficient level of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf (…)"

And this is the new one:

"Providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf (…). This obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual."

So before, you had to make sure your people knew enough about AI. Now you show that you help them build that knowledge, and you don't have to guarantee exactly how much they know. Deployer is the legal term for the organization that uses AI, so if you use an AI tool, that's you.

Lawyers call that an obligation of effort: it's about what you do. The change is in the Digital Omnibus, Regulation (EU) 2026/1744, which entered into force on 27 July 2026.

Who does it apply to?

Article 4 applies to every organization that uses AI, even if you didn't build the tool yourself. It covers your own staff and people who work with the system on your behalf. So look at everyone with access to the admin environment, such as the agency that manages your website or a freelance editor.

Is there a fine?

The AI Act doesn't attach a fixed fine to Article 4, because it isn't in the penalty categories of Article 99. Member states may, however, set their own penalties for infringements of the regulation.

It still counts: if something goes wrong, a supervisor will want to know what you had arranged for the people who work with the system. If you can show that, you're in a stronger position.

What do your people need to know?

A general course on AI is a good foundation, but the law asks for measures that fit the work someone does. If you add an AI search function to your website, for example, everyone needs something different:

• The web editor who manages the sources: the assistant answers from the sources you point it to. If an outdated page is among them, a resident gets an outdated answer.
• The employee who reads the conversations: an AI answer can sound convincing and still be wrong. You use the source references to check whether it's right.
• The agency that builds in the widget: visitors need to see that they're talking to AI, also when they use a screen reader.
• The administrator of the environment: visitors sometimes type personal data into a question. You know who can see the conversations and how long they're kept.

How to arrange it

There's no fixed format you have to follow, but you do need to be able to show that you did it. These four steps take care of that:

• Write down who works with which AI system, including people from outside.
• Decide per role what someone needs to know. An editor who manages sources needs different knowledge than an advisor who reads the analytics.
• Record what you offered and when. For example: "web team, explanation of source management by the supplier, 1 hour, March 2026".
• Look at it again with a new system or after an incident.

A test or a score per employee isn't needed.

How we look at this at vragen.ai

Article 4 applies to us too, for our own people. Beyond that, we want you to have control over the answers, and above all to see how they're formed. That's why the admin environment shows you:

• every conversation, with the sources the AI cited and consulted;
• for each answer, the evaluation scores for reliability, relevance and source relevance;
• the instructions the agent was given, which you can view and adjust yourself;
• the trace: every reasoning step the AI took to reach the answer.

Anyone who follows those conversations and reasoning steps also learns how AI arrives at an answer. How to set up and fine-tune the assistant is in the documentation, for example in best practices and users and roles.

We can't sign off that your organization complies with Article 4, because that obligation lies with you. If you'd like the explanation of vragen.ai for your own team, you can always talk it through with us.

Want to see what that admin environment looks like for you? Create your demo. It's free and reads your own website.

Short answers

Is AI literacy mandatory?

Yes, since 2 February 2025, for every organization that develops or uses AI. So also if you only use an AI tool.

What did the Digital Omnibus change in Article 4?

Since 27 July 2026, you take measures that support the AI knowledge of your people. You don't have to guarantee a specific level for each person.

What's the fine if you don't comply with Article 4?

The AI Act doesn't attach a fixed fine to Article 4. Member states may set their own penalties, though. And it counts if something goes wrong.

Does AI literacy also apply to contractors?

Yes. It also covers people who work with the system on your behalf, such as an agency or a freelance editor with access to the admin environment.

Is a general AI course enough for Article 4?

Not as the only measure. The measures have to fit the work someone does. A general course doesn't tell you what your own assistant does and how to fine-tune it.