AI Literacy (Article 4): What Your Team Must Know Since February 2025

Short answer: it already applies. Since 2 February 2025, Article 4 of the EU AI Act has required that everyone who works with AI on your behalf understands it well enough: what the tool can and can't do, and when a human needs to step in. No exam and no certificate, but demonstrable working knowledge that fits how you actually use AI, and that you put on record. This obligation runs ahead of the better-known chatbot disclosure, which only takes effect on 2 August 2026.

What does AI literacy (Article 4) actually require?

Article 4 requires both providers and deployers of AI systems to ensure, "to their best extent", a sufficient level of AI literacy among their staff and anyone working with AI on their behalf. The bar isn't absolute: the law ties it to the knowledge, experience, and training of your people, and to the context in which you use the AI. In short: AI literacy has been mandatory since 2 February 2025, not at some later date.

This matters for webshops: if you deploy someone else's AI (a chatbot from Shopify, Intercom, or Tidio, or a text or image generator), the law treats you as a "deployer". Deployers fall under Article 4 too. You don't have to build AI to be covered; you only have to use it.

Does Article 4 apply to small businesses too?

Yes. Unlike the European Accessibility Act, the EU AI Act has no exemption for small or micro-enterprises. Where the EAA exempts services from companies under 10 employees and €2 million in turnover, AI literacy applies regardless of your size. What does scale with you is the depth: a team of three with a single support chatbot needs a lighter setup than an agency that builds AI into client products.

So the first question is simple: is any AI running in your organization? We detect which chat widget sits on your site; whether it's AI-driven is something you confirm with your supplier. And if you also use AI for product copy or translations, that counts just as much.

No exam, but provable: what "sufficient" means

There's no test, no diploma, and no mandatory course. The law asks for effort "to your best extent", and in practice that means you can show that your people know what the AI can and can't do, where the risks sit (errors, made-up answers, privacy) and when a human needs to check the work.

The EU AI Office keeps a public overview of how organizations approach this. One caveat: the Commission itself notes that copying someone else's approach wholesale does not automatically put you in the clear. Literacy isn't a checkbox you copy from elsewhere; it's something that fits your own use of AI, and that you put on record.

How to handle AI literacy: light and provable

For most webshops and SME teams, this is something you can sort out in an afternoon:

1. Inventory your AI. Which tools do you use, from which supplier, and for what (chat, copy, images, translations)? This overview is the foundation, and immediately your first piece of evidence.
2. Explain it briefly per role: what the AI does, what people should watch for, and when they shouldn't trust an answer blindly. One clear page per task is enough.
3. Record that it happened: the date, who received the briefing, and what was covered. That's exactly the evidence you'll want to show a supervisory authority.
4. Keep it alive. New tool or new colleague? Add them. Literacy isn't a one-off task but an overview that moves with you.

Step 1 is also where Seviranta's AI transparency check begins: we detect which chat widget runs on your site and record it in an AI register (which AI, what for, status). That register gives you the inventory from step 1 right away, and later your evidence for the transparency obligation.

Article 4 now, Article 50 next

AI literacy is the first obligation that's active, but not the last. From 2 August 2026, your chatbot has to say it's AI at first contact, and AI-generated content has to be marked in a machine-readable way; for generative systems that were already in use before that date, there's a phase-out until 2 December 2026. Breaching those transparency rules carries fines of up to €15 million or 3% of worldwide annual turnover. In each EU country, a national supervisory authority enforces this. We explained earlier how that disclosure works, in our article on whether your chatbot has to say it's AI.

Convenient: the same register that records which AI you use serves as your inventory for Article 4 and your evidence for Article 50. One overview that counts in both cases.

Set up the overview now, while it's still light work. The AI transparency check detects your chat widget, provides the disclosure text, and builds your AI register, for €29 per month (or €290 per year, 2 months free; excl. VAT), standalone or as an add-on to any plan. See the AI transparency check. Haven't run a scan yet? Scan your site for free and see within two minutes which widget is running.

Sources: Article 4, EU AI Act · AI Act implementation timeline · EU AI Office: living repository on AI literacy