EU AI Act AI literacy (Article 4): what your company actually has to do

Most EU AI Act coverage is about the systems at the top of the risk pyramid — what is banned, what is high-risk, what gets fined the most. Article 4 sits at the other end, and it is the one obligation that reaches almost every company that touches AI at all. It is short, it is easy to skim past, and it has been in force since 2 February 2025.

It also changed in July 2026. The Digital Omnibus on AI — Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026 and in force from 27 July 2026 — rewrote Article 4 into a lighter obligation. And since early August 2026, national authorities have been supervising it: the Act’s market-surveillance provisions apply from 2 August 2026, and the Commission’s AI literacy Q&A puts Article 4 supervision at 3 August 2026 onwards.

This is an information service to help you orient, not legal advice. To see which obligations attach to your own AI use, answer three questions.

What Article 4 actually requires

Providers and deployers of AI systems must take measures to support the development of AI literacy among:

• their staff, and
• other people who deal with the operation and use of AI systems on their behalf — which the Commission reads as including contractors, service providers and, in some setups, clients.

The measures have to be proportionate. Article 4 explicitly points to those people's technical knowledge, experience, education and training, the context the AI system is used in, and the people the system is used on.

The key thing most companies get wrong: Article 4 is not tied to the high-risk tier. It is not part of the high-risk regime that the Digital Omnibus pushed to 2 December 2027. If you deploy an AI system — and using an off-the-shelf AI tool in your business generally makes you a deployer — Article 4 applies to you. Risk level changes what is proportionate, not whether you are in scope.

What changed on 27 July 2026

The Digital Omnibus softened the standard. This matters more than it looks:

	Before	After Regulation (EU) 2026/1744
The duty	Take measures to ensure, to their best extent, a sufficient level of AI literacy	Take measures to support the development of AI literacy
Nature	Closer to an obligation of result	An obligation of effort — reasonable measures, not a guaranteed outcome
Individual standard	Ambiguous whether each person had to reach a level	Amended Article 4(1) states the obligation "does not require providers or deployers to guarantee any specific level of AI literacy of any individual"
Support	—	The Commission and Member States must support and facilitate compliance efforts, in particular for SMEs (Art 4(2)); the Commission is to publish practical compliance examples and the AI Board to adopt AI-literacy recommendations (Art 4(3))

What did not change is just as important:

• The obligation was not deleted and not deferred. Unlike the high-risk rules, it has no new future start date — the amended version simply applies from 27 July 2026.
• The scope is unchanged: providers and deployers, staff and those acting on their behalf.

If you had concluded that "everything got delayed" after the Omnibus headlines, this is one of the places that reading breaks down — the same way it does for the transparency rules in Article 50, which landed on 2 August 2026.

The date that changed your exposure

Article 4 has technically applied since 2 February 2025, but it arrived without an enforcement mechanism attached. That is what changed in August 2026.

The European Commission's Q&A on AI literacy states that the supervision and enforcement rules apply from 3 August 2026 onwards — the point at which national market surveillance authorities take up their supervisory role. So the practical position is:

• 2 February 2025 — the obligation applies. In force, but with no authority actively supervising it.
• 27 July 2026 — the amended, softer version of Article 4 applies.
• 3 August 2026 — supervision and enforcement of Article 4 begins at national level (the date given in the Commission’s AI literacy Q&A; the Act’s market-surveillance provisions apply from 2 August 2026).

You have had the duty for over a year and a half. Since August 2026 there has been someone empowered to ask about it.

What counts as compliance

There is no certification, no mandated curriculum, and no minimum number of training hours. The Commission's Q&A is direct about this: there is no one-size-fits-all approach to AI literacy, and no strict requirements or mandatory trainings are imposed — the question of whether training is mandatory is covered in depth in Is AI training mandatory under the EU AI Act?

That cuts both ways. Nobody can sell you the compliant course — but "we sent everyone a link to a webinar" is equally not the point. What a proportionate programme tends to look like:

1. Know where AI is in use. You cannot size a literacy measure without knowing which teams use which tools, and for what. This inventory does double duty — it is the same list Article 50 and any high-risk assessment depend on.
2. Segment by role. A support agent using an AI reply drafter, an engineer integrating a model, and an HR lead running a screening tool need genuinely different things. Proportionality is the standard, and role is the cleanest way to apply it.
3. Cover the risks specific to your use. Not general AI theory — the failure modes of the tools you actually run: hallucination, bias, over-reliance on output, what must never be pasted into a prompt, when a human has to decide.
4. Include people who are not employees. Contractors and service providers operating AI on your behalf are in scope.
5. Write it down. The obligation is now one of effort, which means the evidence is the effort: what you did, for whom, when, and why it was proportionate. An undocumented measure is difficult to demonstrate to an authority that asks.
6. Revisit it when things change. New tool, new use case, new regulation — the measures should move with them.

The Commission also maintains a living repository of real AI literacy practices — initially collected from AI Pact signatories and since opened to other organisations — which is a reasonable place to calibrate what peers are doing without inventing a standard from scratch.

How this fits the rest of your AI Act exposure

Article 4 is rarely the whole picture, and it is the cheapest of the obligations to address. For most companies the near-term stack looks like this:

Obligation	Applies from	Tied to high-risk?
Article 4 — AI literacy	2 Feb 2025 (supervision from 3 Aug 2026)	No
Article 5 — prohibited practices	2 Feb 2025 — two prohibitions added by the Omnibus apply from 2 Dec 2026	No
Article 50 — transparency	2 Aug 2026	No
Annex III high-risk obligations	2 Dec 2027	Yes
High-risk embedded in regulated products	2 Aug 2028	Yes

Three of the five reach companies with no high-risk system at all — and the first three are already live or nearly so. If you have not mapped your position, start with which EU AI Act obligations apply to your company, and confirm your high-risk position separately with is your AI system high-risk?.

What to do now that supervision has started

Realistically, this is a week of work, not a quarter:

1. Inventory every AI tool in use and who uses it.
2. Group users by role and decide what each group needs to understand.
3. Deliver something proportionate — a short internal session, a written guide, tool-specific guidance — and record who received it.
4. Note the people acting on your behalf who are not employees, and cover them too.
5. Diarise a review for when your tools or the rules change.

Article 4 will not produce the AI Act's headline fines. But it is the obligation an authority can ask about first, it applies whether or not you have anything high-risk, and since August 2026 there is somebody positioned to ask.

To see the obligations that match your own AI use, sorted by deadline, answer three questions. To be told when any of them change — as Article 4 did in July — join the waitlist.

The official text is Regulation (EU) 2024/1689; AI literacy is Article 4, as amended by Regulation (EU) 2026/1744. This article is an information service to help you orient — it is not legal advice.