AI Literacy Training and the EU AI Act: What Article 4 Requires Before August 2026

EU AI Act Article 4 makes AI literacy training a legal duty for companies using AI: what it requires, who it covers, the August 2026 deadline, how to meet it.

If your company uses AI at work, you already have a training obligation under the EU AI Act. It is called AI literacy, it sits in Article 4, and it has applied since 2 February 2025. From 2 August 2026, national authorities can enforce it. Most organizations are not ready.

TL;DR: Article 4 of the EU AI Act requires providers and deployers of AI to make sure their staff have a "sufficient level of AI literacy." Almost every company is a deployer. The rule is already in force, national enforcement starts 2 August 2026, and the 2026 Digital Omnibus did not delay it. There is no fixed curriculum and no standalone fine, but you are expected to run role-appropriate AI literacy training and keep a record that people completed it.

A quick, honest caveat before we go further: this article explains the obligation in plain terms, it is not legal advice. Check your specific situation with your own counsel or DPO.

What is EU AI Act Article 4?

Article 4 is the AI literacy provision of the EU AI Act (Regulation 2024/1689). It says 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." You can read the full Article 4 text here.

In plain words: if people at your company use AI to do their jobs, you have to help them understand what they are using and where it can quietly go wrong. It applies whatever risk tier your AI falls into, which is what makes it the broadest single duty in the whole Act.

Who has to comply with the AI literacy rule?

Two groups: providers (organizations that build or place AI systems on the market) and deployers (organizations that use AI systems under their own authority). The European Commission's own AI literacy Q&A is clear that the duty covers both.

Here is the part people miss, and I run into it constantly. Most companies are deployers, not providers. If your teams use ChatGPT, Copilot, an AI note-taker, an AI hiring tool, or an AI feature inside software you bought, you are a deployer, and Article 4 applies to you. It also reaches people who operate AI on your behalf, so contractors and agency staff count too. There is no small-business carve-out and no "we only use one tool" exemption.

When does Article 4 apply, and when is it enforced?

The obligation has applied since 2 February 2025. That date has passed. So on paper, the duty is live right now.

What changes on 2 August 2026 is enforcement. National market-surveillance authorities (not the EU AI Office) gain the power to supervise and act on Article 4 from that date. The Commission Q&A states this directly: the AI Office does not police Article 4, national authorities do, and their powers apply from August 2026. Member States were meant to name those authorities and set penalty rules the year before.

So the honest framing: the obligation is already here, and the people who can act on it get their teeth in August 2026.

Did the Digital Omnibus delay the AI literacy deadline?

No. This is the single most common misunderstanding right now, so it is worth being precise.

The EU adopted the Digital Omnibus on 29 June 2026 (the Council gave the final green light here). It pushed back the deadlines for high-risk AI systems: the Annex III obligations moved to 2 December 2027, and the Annex I embedded-product rules moved to 2 August 2028.

It did not touch Article 4. AI literacy stays on the 2 August 2026 timeline, and so do the Article 50 transparency rules (telling people when they are dealing with AI). The two duties that reach almost every organization are exactly the two that were left in place. If anything, the Omnibus made the near-term picture simpler: high-risk work has more room, but AI literacy and transparency are still due this summer.

What does a "sufficient level of AI literacy" actually require?

The law does not hand you a syllabus. The wording is "to their best extent" and "sufficient," which is deliberately flexible. The Commission says it will not impose one-size-fits-all requirements, and that the right level depends on people's roles, their technical background, and the context in which they use AI.

In practice a defensible program tends to cover three things:

• A shared baseline for everyone who touches AI: what these systems are, what they are good and bad at, and where they can mislead you.
• Role-aware content: a recruiter using an AI screening tool, a support agent using an AI assistant, and an engineer wiring a model into production need different depth.
• Awareness of the specific systems you actually use, including their known risks and your internal rules for them.

The Commission also keeps a public living repository of AI literacy practices you can look at for real examples. One caution it makes explicit: copying someone else's program gives you no presumption of compliance. Your training has to fit your context.

Does Article 4 come with a fine?

Not directly. Article 4 is not listed in the AI Act's penalty article (Article 99), so there is no dedicated fine attached to it. It is enforced indirectly: from August 2026 a national authority can require corrective measures, and weak AI literacy can count against you when something else goes wrong, for example after an AI-related incident. Any actual penalties after that date flow from national law, which varies by country.

One thing to get right in your own messaging and planning: do not quote the headline 35 million euro or 7 percent of turnover figure for this. That top tier is for prohibited practices, a different part of the Act. Mixing up the tiers is the fastest way to lose credibility with a compliance team. The realistic reason to act on Article 4 is readiness and documented good faith, not a specific fine.

Does AI literacy training need a certificate?

No. Article 4 does not require an exam, an accredited certification, or a specific credential. Anyone claiming their badge equals "AI Act compliance" is overselling.

What actually matters to a regulator or an internal audit is evidence: a documented, role-appropriate program, and a record showing which named people completed which training and when. A certificate of completion is fine and useful as that record. An accredited "certification" is a different thing and you do not need one to meet the duty. Build the completion record as a first-class part of the program, because that audit trail is the thing you will be asked for.

How to build AI literacy training that meets Article 4

A workable program in five steps:

1. Inventory where AI is actually used. List the tools and AI features in play across teams, including the ones people adopted without asking. You cannot train for what you have not mapped.
2. Segment by role. Group staff by how they use AI, from light everyday users to people building or deploying systems. This is what "appropriate to context" means in practice.
3. Make the content fit your reality. Generic modules teach the Act in the abstract. Better training uses your own AI policies, your approved tools, and your real use cases, so people learn what applies to their day.
4. Capture proof of completion. Track who finished what, with dates, in a form you can export for an audit.
5. Keep it current. When you adopt a new tool, or the rules change, refresh the training. AI literacy is not a one-time event.

Common mistakes to avoid

• The one-off webinar. A single all-hands session with no record and no follow-up is the classic box-ticking move, and it is exactly what "to their best extent" is meant to rule out.
• No completion trail. If you cannot show who was trained, you cannot show you met the duty.
• Off-the-shelf content that ignores your tools. Training that never mentions the AI your people actually use is hard to defend as context-appropriate.
• Overclaiming. Promising "guaranteed compliance" or an "official certificate" is the kind of thing a sharp DPO will notice, and it undermines trust.

How Uncoursed helps

The Commission's own standard, literacy that is "appropriate to context," is the hard part of doing this well. Off-the-shelf courses teach the regulation in general. They cannot teach it against your policies and your tools.

Uncoursed turns a dense document into a real interactive course. Give it the AI Act text, or your own AI policy and tool guides, and it builds a structured course with lessons, quizzes, an exam, and spaced-repetition flashcards drawn from your material, plus a learning buddy that answers only from your sources. When the Act does not cover something, the learning buddy says so instead of guessing, which is the honest way to run AI over a legal text. You also get a completion record per learner, the audit trail Article 4 practically requires.

We built a free EU AI Act course you can go through to see the format. If you want AI literacy training shaped around your own tools and policies, that is the same engine pointed at your material. And it is the same thing we do for corporate training and onboarding generally: take the knowledge you already have and turn it into something people finish and remember.

This is general information about Article 4, not legal advice. Validate your specific obligations with qualified counsel or your DPO.

FAQ

Is AI literacy training mandatory under the EU AI Act?

Yes. Article 4 requires providers and deployers of AI to ensure a sufficient level of AI literacy among staff and anyone using AI on their behalf. It has applied since 2 February 2025.

When does the AI Act AI literacy requirement get enforced?

National authorities can enforce Article 4 from 2 August 2026. The obligation itself has been in force since February 2025. The 2026 Digital Omnibus delayed high-risk deadlines to 2027 and 2028 but did not delay Article 4.

Which companies must provide AI literacy under the AI Act?

Almost all of them. Any organization that uses AI systems at work is a deployer and falls under Article 4, regardless of size or the risk level of the AI. It also covers contractors who use AI on your behalf.

Is there a fine for skipping Article 4?

There is no standalone fine for Article 4. It is enforced indirectly through national authorities from August 2026, and penalties come from national law. The 35 million euro tier applies to prohibited practices, not to AI literacy.

Do you need a certificate to comply with Article 4?

No. There is no required certificate or exam. What matters is a role-appropriate program and a record of who completed it, which you can keep as evidence of meeting the obligation.

For more on why passive training fails and what makes it stick, see our guides on why employees forget most training and using AI across a corporate learning program.

New posts, in your inbox.

Memory science, product field notes, and occasional opinions about the state of online learning.