Is AI training mandatory under the EU AI Act?

If a vendor has told you the EU AI Act makes AI training mandatory, they are selling you something. The honest answer is more useful — and more reassuring: the Act does not require a specific training course, a certificate, or a guaranteed level of knowledge. It requires something softer and more practical. Here is what the law actually says.

This is an information service to help you orient, not legal advice. For the obligations tied to your own AI use, see which apply to your company.

What Article 4 actually requires

AI literacy lives in Article 4 of the AI Act. It applies to every provider and deployer — at every risk tier, not just high-risk AI — which makes it the broadest single duty in the Regulation.

Crucially, it has never said "run a training course." Even in its original form it asked you to ensure a sufficient level of AI literacy among your staff and anyone operating AI on your behalf — with the level appropriate to their role, your context, and the risk of the systems. Training is one route. So are role-appropriate briefings, written guidance, guardrails, and good tooling.

The Digital Omnibus made it even softer

Here is the part most guides haven't caught up to. Regulation (EU) 2026/1744 — the Digital Omnibus on AI — was published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026, amending the AI Act. It re-wrote Article 4.

The duty changed from ensuring a sufficient level of AI literacy to taking measures to support the development of it — an obligation of effort, not of result. And it added a sentence that removes any doubt:

This obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual.

In plain terms: you are expected to make a genuine, proportionate effort to help your people understand the AI they use — not to certify that each person has hit some standard.

So what do you actually have to do?

"Not mandatory" is not the same as "nothing." The practical requirement is to act, proportionately, and be able to show it:

• Match the effort to the risk and the role. A team shipping a high-risk hiring model needs more than a team using an internal chatbot.
• Pick the measures that fit — a short internal session, a one-page policy, onboarding notes, usage guardrails. Any of these can satisfy Article 4; none is required over the others.
• Write it down. The value of documentation isn't the paperwork — it's that "we took proportionate measures to support AI literacy" is a claim you can actually back up if a regulator asks.

Why the "mandatory training" framing is now doubly wrong

A lot of content still cites the pre-amendment Article 4 — the "ensure a sufficient level" wording — and reads it as a hard training mandate. As of 27 July 2026 that wording is out of date on two counts: it was never a training mandate, and the obligation is now explicitly one of effort. If your compliance plan was built on "we must train everyone to a standard," it is over-scoped.

The fastest way to see which obligations — literacy and otherwise — actually apply to you: answer three questions about your AI use. To be told the moment a rule like this changes again, join the waitlist.

The official texts are Regulation (EU) 2024/1689 (Article 4) and Regulation (EU) 2026/1744 (the Digital Omnibus amendment). This article is an information service to help you orient — it is not legal advice.