AI literacy under the EU AI Act: what Article 4 asks of you

In short: AI literacy is a duty in Article 4 of the EU AI Act. Since 2 February 2025, every organisation that builds or uses AI must take steps so its people know enough to use it well. There’s no set course or certificate. What counts is that people know what their tools can do, where they fail and what to check.

You’ve given your team AI tools so they can work faster and serve customers better. That only works when people know what the tool is good at and where it slips. The EU AI Act now asks for exactly that. AI literacy is one of the first duties that already applies. And it covers almost every organisation that uses AI, not only the ones with high-risk systems.

What does Article 4 of the EU AI Act ask?

Article 4 asks providers and deployers of AI systems to take measures to ensure, to their best extent, a sufficient level of AI literacy. That covers their staff and other people who operate or use AI on their behalf. The law says to take into account people’s knowledge, experience, education and training. It also says to look at the context the AI is used in and the people it affects. Article 3 defines AI literacy as the skills, knowledge and understanding to use AI in an informed way. That includes knowing its opportunities, its risks and the harm it can cause. The duty has applied since 2 February 2025. So this isn’t a future task. If your people use AI at work today, it applies today.

Who does the AI literacy duty apply to?

It applies to every provider and every deployer, whatever the risk level of the AI. A provider builds an AI system or puts its name on it. A deployer uses an AI system in its work. If you’re not sure which you are, our post on provider or deployer helps you work it out.

That means a marketing team using a writing tool is in scope. So is a support team with a chatbot, and an HR team that uses AI to sort CVs. It also covers people who aren’t on your payroll but use AI for you, such as contractors or a call centre partner. The level you need does depend on the risk. A team that drafts social posts needs less than a team whose AI helps decide on loans.

What should your people actually know?

Your people should know the specific tools they use, not AI in general. A course on how large language models work is a fine start. But it won’t tell a support agent that your chatbot gets refund rules wrong for orders from abroad. Our advice is to cover four points for each tool and each group of users.

  1. What it’s for. The tasks the tool may be used for, and the tasks it may not.
  2. Where it fails. The kinds of mistakes it makes, such as made-up facts, outdated answers or a different tone for different people.
  3. What to check. Which outputs a person must review before they reach a customer or a decision.
  4. When to stop. What to do when the tool gets something wrong, and who to tell.

Keep it short and practical. One page per tool that people actually read beats a long training module they click through once.

How do you show you’ve done enough?

You show it with records, not with a certificate. The European Commission’s questions and answers on AI literacy say no certificate is needed. Organisations can keep an internal record of what they did instead. National authorities supervise the duty, and they’ll look at what you did in relation to your risks.

A useful record has three parts. A list of the AI tools in use and who uses them. What each group was told or trained on, and when. And how you keep it up to date as tools change. There’s one point to watch. In November 2025 the European Commission proposed changes to the AI Act in its Digital Omnibus, including to how the AI literacy duty works. Check the current text before you plan for the long term. Whatever happens, staff who understand their tools make fewer costly mistakes.

Why your own test results make the best training material

General examples of AI mistakes are easy to dismiss. “Our tool wouldn’t do that.” Real answers from your own tool are not. Say people see that their own chatbot gave a wrong price or leaked part of its instructions. Then they understand at once what to look out for.

That’s why we’d build AI literacy on test results. Run your tools through a fixed set of test questions, keep the answers that went wrong and use them in the one-pager for each tool. Then repeat it. Even well-designed AI systems drift and misbehave over time, so what people learned in spring can be out of date by autumn. A vendor update can change behaviour overnight, as our post on model drift explains.

There’s a trade-off. Testing every tool in depth isn’t worth it for low-risk use. For a writing tool, a short list of known weak spots may be enough. Save the deeper tests for AI that talks to customers or affects decisions.

What this means for your team

Start this week with a list: every AI tool in use, who uses it and what for. Pick the tool that reaches the most people and write its one-pager with the four points above. Note the date and who received it. That’s your first record. For the bigger picture of what the law asks, see our page on EU AI Act compliance.

The hard part is keeping the “where it fails” section true over time. VeritasAI’s PRO plan runs automated weekly tests mapped to the EU AI Act, ISO 42001, OWASP and NIST standards. You can review the exact responses your model gave during the tests. That gives you real examples for your team and a dated record that you’re watching your AI.

Show your team what your AI really does, every week.

See the PRO planBook your demo

Frequently asked questions

Is AI literacy training mandatory under the EU AI Act?

Taking measures is mandatory for providers and deployers. Training is one way to do it, but the law doesn’t prescribe a course. Guidance, instructions and hands-on sessions can count too.

Is there a fine for not meeting the AI literacy duty?

The AI Act doesn’t list a specific fine for Article 4 in its fine levels. National rules decide how it’s enforced. A lack of AI literacy can still count against you when something goes wrong with your AI.

Does AI literacy apply if we only use ChatGPT or Copilot?

Yes. Using a general AI tool at work makes you a deployer. Your people should know what the tool is allowed for and where it makes mistakes.

How often should AI literacy be refreshed?

Whenever a tool, its use or its behaviour changes, and at least once a year. New staff should get it when they start using the tool.

Scroll to Top