EU AI Act Article 4: What the AI Literacy Duty Actually Requires
A plain-English guide to Article 4 of the EU AI Act — who it applies to, what "reasonable measures" actually means, and what changes from 2 August 2026.
If your team uses ChatGPT, Copilot, or any other AI tool at work, there's a good chance a piece of EU law already applies to you — even if nobody at your company has heard of it yet.
Article 4 of the EU AI Act is the "AI literacy" duty. It has been in force since February 2025, and from 2 August 2026, national regulators gain formal power to supervise and enforce it. Unlike most of the AI Act's headline-grabbing rules, this one was not pushed back to 2027 in the 2026 Digital Omnibus revision — it's live now.
Who does Article 4 apply to?
Almost everyone. The obligation isn't limited to companies that build AI systems — it covers any organization whose staff use AI tools as part of their work. That includes ordinary, everyday use: drafting emails with ChatGPT, summarizing documents with Copilot, or using an AI assistant for research. There's no small-business exemption.
If people at your company are using AI tools — even informally — Article 4 is already relevant to you.
What does it actually ask you to do?
This is where a lot of the online commentary overstates things. Article 4 does not require you to hire a certified AI compliance officer, buy a specific accredited course, or pass an exam. Following the Digital Omnibus revision, it's an effort duty, not an outcome duty — the law asks organizations to "take measures to support the development of" AI literacy, proportionate to the context and risk involved.
In practice, that means:
- Map where AI is actually used in your organization, and by whom.
- Match training to role and risk — someone drafting client emails with AI needs different guidance than someone reviewing AI-assisted financial analysis.
- Deliver genuine, practical training — not a box-ticking exercise nobody remembers.
- Keep a record of it. Documentation is what regulators will actually look for.
There is no required course, certifying body, or format. A certificate is a convenient, defensible way to document that reasonable steps were taken — it isn't the only way, and it isn't legally mandated. Any source claiming otherwise is overstating the requirement.
What changes on 2 August 2026?
Nothing about the underlying duty changes — it's applied since February 2025. What changes is enforcement: from 2 August, national supervisory authorities gain the formal power to check whether organizations are actually meeting it. That's the date that turns "a rule that technically exists" into "a rule someone might actually ask you about."
A note on penalties: because enforcement powers only start in August, and penalty regimes vary by member state and by which part of the AI Act is being enforced, don't trust any specific fine figure quoted for Article 4 literacy failures specifically — most numbers circulating online are pulled from the AI Act's general penalty tiers for unrelated violations, not this duty.
The practical question to ask this week
Does your team have any documented AI training right now? Not a perfect program — just something real, and something written down.
If the honest answer is no, that's the gap worth closing before enforcement begins — not because of a specific fine, but because "we did nothing" is a difficult position to defend if it's ever asked about.
Collabr.ai is a hands-on, chat-based course built for exactly this: practical exercises — spotting hallucinations, protecting confidential information, knowing what never to paste into an AI tool — that give your team real judgment, and give you a completion certificate for your files. See how it works.