AI Literacy Under Article 4: The EU AI Act Duty in Force

If your company uses AI, the first obligation under the EU AI Act is not a 2026 or 2027 deadline. It has already been binding since 2 February 2025. Article 4 requires you to make sure the people who use AI on your behalf are AI-literate. There is no certificate to buy and no exam to sit. But there is a result you must reach and a paper trail you should be able to produce. Here is what the duty actually says, who it covers, how to run the training, and how to prove you did.

Quick Answer. Yes. EU AI Act Article 4 requires every provider and deployer to ensure a sufficient level of AI literacy among staff and anyone using AI on their behalf. It is not a future deadline: the duty has applied since 2 February 2025. It is an obligation of result you must evidence, not a certificate.

Last updated: 13 July 2026 · Verified against Regulation (EU) 2024/1689 as amended by the Digital Omnibus (adopted and signed; awaiting Official Journal publication).

Summary

AI literacy under Article 4 — a duty already in force, and how to prove you met it
│
├─ It already binds you
│   ├─ In force since 2 Feb 2025 — not a future deadline
│   └─ Penalties live since 2 Aug 2025; the Omnibus left Art. 4 untouched
│
├─ What Article 4 actually demands
│   ├─ Providers AND deployers ensure their people are AI-literate
│   ├─ An obligation of result — no exam, no certificate, no set course
│   └─ Scale the level to each person's role, risk and context
│
└─ Do it — then evidence it
    ├─ Map who touches AI · set the level · deliver the training
    ├─ Record who, what and when — the evidence trail is the point
    └─ Refresh as tools change and keep the log for a regulator

What does Article 4 of the EU AI Act require?

Article 4 is short, and it is worth reading in full. So much guidance paraphrases it into something narrower than it is:

"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, taking into account their technical knowledge, experience, education and training and the context the AI systems are to be used in, and considering the persons or groups of persons on whom the AI systems are to be used." ¹

In plain terms, that is one obligation with several moving parts. You must take measures: active steps, not a policy you never put into practice. The target is a sufficient level of AI literacy, not a fixed syllabus. The bar moves with the situation. It binds both providers (those who build or supply AI systems) and deployers (any organisation using an AI system under its own authority). And it reaches beyond your payroll to other persons dealing with the operation and use of your AI. That means the contractors, agency staff and outsourced operators who touch the system on your behalf.

Article 3 of the same regulation defines AI literacy as the skills, knowledge and understanding that let people deploy AI systems appropriately and be aware of the opportunities, risks and possible harm ¹. So "literacy" here is practical competence for the systems your people actually use, not a general awareness course about robots. For how this duty sits inside the wider set of records a company using AI has to keep, work through our complete EU AI Act documentation guide. It maps Article 4 against the rest of the pack.

Since when the duty applies

Since 2 February 2025. Article 4 became applicable on that date under Article 113 of the regulation, in the same first wave as the Article 5 prohibitions ^1,3. That is the single fact most companies get wrong. The AI-literacy duty is not something arriving with the transparency or high-risk deadlines. It is already in force, and has been for over a year.

Two consequences follow. First, if you have not run any AI-literacy measures, you are not "getting ready" for a deadline. You are already behind one. Second, the exposure is not theoretical. The Article 99 penalty regime has applied since 2 August 2025, so the enforcement machinery around the Act is live ^1,3.

It is also worth killing a common piece of misinformation. The Digital Omnibus, the simplification amendment that moved parts of the Act, did not touch Article 4. It shifted the Annex III high-risk obligations to 2 December 2027 and the regulatory sandboxes to 2 August 2027. It left the literacy duty exactly where it was. If you want the corrected schedule in full, our corrected post-Omnibus deadline timeline lays out what moved and what did not. (Verified on 13 July 2026.)

Who must be trained — and to what level?

Everyone who "deals with the operation and use" of your AI systems is in scope. But the level of literacy is deliberately relative. Article 4 tells you to take into account each person's technical knowledge, experience, education and training, and the context and risk of the systems they use ¹. A colleague pasting text into a drafting assistant does not need the same depth as the person who selected, configured and monitors a model that screens suppliers.

The practical move is to start from your inventory of AI systems and the roles that touch each one. If you have not mapped that yet, start with our guide on how to build your AI register in about 90 minutes. It gives you the who-and-what that this duty scales against. Then set a proportionate level per role:

The point of the table is not to certify anyone. It is to show, role by role, that you thought about what "sufficient" means for the systems those people actually operate. That is exactly what Article 4 asks you to weigh ^1,2.

Running AI-literacy training

Article 4 is an obligation of result rather than a prescribed course, so you have wide latitude on format. You should use it to keep the effort proportionate. The Commission's own AI-literacy guidance is explicit that there is no single mandated programme. Organisations are expected to design measures that fit their staff and their use of AI ². In practice, a defensible programme has five moving parts, and they run as a pipeline rather than a one-off event:

In words, the programme works like this:

1. Map who deals with AI on your behalf: staff, contractors, agency operators and any vendor running a system for you. This is the population Article 4 covers, and it is wider than your headcount ¹.
2. Set the level per role, scaled to risk and context using the table above. A low-risk drafting aid warrants a short briefing; a system that affects people warrants real depth.
3. Deliver the training in whatever format fits: a live briefing, a short e-learning module, written role guides, an acceptable-use policy staff sign, or supplier-provided sessions for a tool you bought. Content should cover what the system does, its limits, the guardrails and how to escalate.
4. Record who was trained, on what, and when: the step most organisations skip, and the one that turns "we did some training" into evidence.
5. Refresh it when you add tools, change use cases or onboard people. "Sufficient" is a moving target, so a static 2025 slide deck stops being sufficient the moment your AI estate changes.

You do not need a bespoke platform to do this. If you would rather generate the programme and the records than build them from a blank page, the free EU AI Act Documentation tool produces both. They are an AI-literacy programme and a training-record template, editable documents you can adapt and keep on file.

Generate your own EU AI Act report + document pack — free, in ~20 minutes: https://aiprioritymap.com/en/tools/eu-ai-act

Documenting the training: the evidence trail

The duty is an obligation of result, but nobody can see a result. They can only see the record of it. So the practical test of Article 4 compliance is whether, when a customer's procurement team or a regulator asks, you can produce a paper trail showing you took proportionate measures. The Commission frames Article 4 as an ongoing obligation to keep under review. That only works if you are recording what you did ².

A good AI-literacy evidence trail has two artefacts. The first is a short AI-literacy programme document that states your approach. It covers which roles are in scope, what level each needs, what content you cover and how often you refresh. The second is a training record that logs the delivery. At minimum, a defensible record captures:

Keep both alongside your AI register so the whole picture sits in one place. The difference between a checker and a document generator shows here. Free readiness checkers tell you whether the Act concerns you, but the EU AI Act Documentation tool approach generates the actual programme and training-record templates you file. Those are the artefacts rather than the diagnosis alone. It is a documentation aid, not legal advice, an audit or a certification.

What happens if you skip it

There is no line-item "AI-literacy fine" in the Act. That lulls some organisations into treating Article 4 as optional. It is not. The duty is legally binding and has been since February 2025. The Article 99 penalty regime has applied since August 2025, and national authorities supervising the Act came online in 2026 ^1,3. A regulator assessing any AI-related complaint can reasonably ask what steps you took to make your operators competent, and "none, on record" is a poor answer.

The bigger day-to-day exposure is commercial and legal rather than a direct fine. Enterprise procurement questionnaires increasingly ask suppliers to evidence AI governance, and an AI-literacy programme is one of the cheapest boxes to be able to tick. And if an AI system your staff misused causes harm, the absence of any literacy measure weakens your position. AI use rarely sits alone: it usually processes personal data too. So the literacy duty overlaps with your existing obligations. Our guide on how the AI Act sits alongside GDPR and UK data rules shows where the two regimes meet. The controls reinforce each other, and the evidence trail does double duty.

Related insights

• EU AI Act Compliance: The Documentation Guide — the deployer-first path and the document pack this literacy duty sits inside.
• EU AI Act Deadlines in 2026: What the Omnibus Changed — the corrected timeline showing Article 4 was never delayed.
• How to Build Your AI Register in About 90 Minutes — the inventory of who and what touches AI that Article 4 scales against.

Last updated: July 2026. Version 1.0.