Skip to main content
DEFINITION

What is AI literacy?

The first AI Act obligation to take effect, and the least known.

The AI Act is known for risk classes and fines. But the first obligation to take effect is about your people: make sure everyone working with AI understands enough about it.

Definition Four roles Making it demonstrable

AI literacy is people’s ability to use AI systems with understanding and judgement: knowing what such a system can do, recognising when an answer is wrong, and understanding what it must not decide. Article 4 of the EU AI Act has obliged organisations since 2 February 2025 to ensure this for everyone working with AI on their behalf, in proportion to each person’s role, experience and the context the system is used in. No exam and no register is prescribed, but the effort must be demonstrable. W69 AI Consultancy in Amstelveen sets this up and keeps it demonstrable.

2 Feb
2025: Article 4 in force
2
roles: provider and deployer
0
mandatory exams or certificates
16 July
2026: we are demonstrably literate

Regulation (EU) 2024/1689, the AI Act, Article 4. This obligation has applied since 2 February 2025 and prescribes no exam or register. This article is explanation, not legal advice; have your own situation reviewed.

THE PROBLEM

The obligation almost nobody knows about

The AI Act is known for risk classes and fines. But the first obligation to take effect is about none of that: it is about your people. Since 2 February 2025 you have to make sure everyone working with AI understands enough about it.

It applies to you too

Article 4 addresses providers as well as organisations that merely use AI. The latter is almost everyone, including those who only bought software with AI inside it. You do not have to build anything to fall under it.

Sufficient is not the same for everyone

The law deliberately does not say how much knowledge is needed. It depends on someone’s role, experience and training, and on what the system is used for. A receptionist and a board member both need something, but not the same thing.

It is not an IT training

Most organisations book a prompting course and consider themselves done. What matters is judgement: recognising when an answer is wrong, knowing what a system may not decide, and being able to explain what it does.

WHO AND HOW MUCH

Four roles, four levels

The depth differs per role. What does not differ is that everyone working with the system has to understand something about it.

Who needs to be AI literate, and how deeply At the top the obligation from Article 4 of the EU AI Act. Below it four roles, each with a bar showing how deep the knowledge has to go, from everyday users up to the board. At the bottom, that sufficient is not the same for everyone, and that there is no exam but it does have to be demonstrable. ARTICLE 4, EU AI ACT, IN FORCE SINCE 2 FEBRUARY 2025 Ensure your people are sufficiently AI literate, by role and by context WHO WHAT THEY MUST BE ABLE TO DO Everyday users Recognise when an answer is wrong Process owners Judge where AI may and may not decide Procurement and legal Ask a supplier the right questions The board Account for what the system does Sufficient is not the same for everyone: it depends on role, experience and risk No exam, no register, no mandatory certificate. But it must be demonstrable.
OUR APPROACH

Demonstrably literate in four steps

There is no exam and no register. What a regulator or a buyer does ask is: show us you thought about it and that it happened.

1

Map who works with AI

Not only the people who chose it themselves, but everyone working with bought-in software that has AI inside. That list is almost always longer than expected and is the basis for everything that follows.

2

Define what sufficient means per role

Write down per group what someone must be able to do. For one that is recognising a wrong answer, for another it is being able to account for it. This is the step organisations skip, and precisely the step the law asks of you.

3

Arrange it and repeat it

One session at onboarding is not literacy. Models change, your own systems change and people move roles. Record when you repeat it and why that rhythm fits your risk.

4

Make it demonstrable

Who did what, when, and what did it cover. Without that trail your effort cannot be distinguished from no effort. It is also exactly what a buyer at a healthcare, finance or government client will ask for.

We are demonstrably AI literate ourselves. Denis Wortelboer completed the AI Literacy course at BAASIN on 16 July 2026, and thereby demonstrably holds the competence to deploy AI responsibly and critically in a professional context. That is not a seal of approval and it is not mandatory either, but it is the first thing we could demonstrate about ourselves before setting it up for anyone else.

W69 AI Consultancy maps who works with AI at your organisation, defines what sufficient means per role, arranges it and keeps it demonstrable. One supplier, including when a buyer asks about it later.

FREQUENTLY ASKED

What people ask about this

Article 4 of the EU AI Act obliges organisations to ensure that their staff, and others working with AI systems on their behalf, understand enough about them. What counts as sufficient depends on that person’s role, experience and training, on what the system is used for and on whom it is used. The obligation has applied since 2 February 2025.

Yes. Article 4 addresses not only providers of AI but also organisations that deploy it. The latter is almost everyone, including those who only encounter AI inside software they bought. You do not have to build anything yourself to fall under it.

No. The AI Act prescribes no exam, no register and no mandatory certificate. What counts is that you can show you thought about it and that it happened: who did what, when, and what it covered. Without that trail your effort cannot be distinguished from no effort.

Usually not. Prompting is a skill; literacy is judgement: recognising when an answer is wrong, knowing what a system may not decide, and being able to explain what it does. A course that only covers operation does not cover the obligation.

The practical consequences usually arrive before any fine. Buyers at healthcare, finance and government clients now ask about it, and in an incident the question of whether your people knew what they were doing becomes immediately relevant to liability. Have your own situation reviewed legally; this article is explanation, not advice.

The law names no interval. What works is a rhythm matching how much changes: the models, your own systems and people’s roles. Record that rhythm and explain why it fits your risk. One session at onboarding is demonstrably too little.

NEXT STEP

Do you know who works with AI at your organisation?

The AI Navigator™ maps where your organisation stands, including who works with AI and what Article 4 asks of you as a result.

RELATED

Go deeper

Home Services AI Scan Sectors WhatsApp