Until 27 July 2026
Ensuring a sufficient level
The original text asked organizations to ensure, as far as possible, a sufficient level of AI literacy among their staff and others who work with AI on their behalf.
AI literacy helps your people work faster and more safely with AI. Here you can read what article 4 of the EU AI Act asks of your organization after the change of 27 July 2026, what works in practice and how to show which measures you took.

AI literacy means your people know what AI can and cannot do, how to use it responsibly and what their own role is. An AI-literate colleague knows why a language model is sometimes convincingly wrong, which data does not belong in a chatbot and when to double-check an outcome.
For organizations, the AI Act, the European AI regulation, also covers it. Article 4 asks you to take measures for the AI literacy of everyone who works with AI on your behalf. Which measures fit depends on your people and on how you use AI.
Article 4 of the AI Act is about AI literacy. It has applied since 2 February 2025 to organizations that provide or use AI. On 27 July 2026, the article was amended by the Digital Omnibus on AI, Regulation (EU) 2026/1744.


Until 27 July 2026
The original text asked organizations to ensure, as far as possible, a sufficient level of AI literacy among their staff and others who work with AI on their behalf.
Since 27 July 2026
The new text asks organizations to take measures to support the development of AI literacy. They take into account the knowledge, experience and training of their people, the context and the people on whom the AI is used.
Providers and deployers do not have to ensure a specific level of AI literacy for each person. The new text says so explicitly.
Source: Regulation (EU) 2026/1744, article 1, point 5, on EUR-Lex (opens in a new tab)
Yes: taking measures is required. The law leaves the form up to you. Below you see per topic what the law asks and what we advise.
| Topic | What the law asks | What we advise | Source |
|---|---|---|---|
| Measures for AI literacy | Required since 2 February 2025, for everyone who works with AI on your behalf. | First map which AI you use and who works with it. | AI Act, article 4 |
| Level for each person | No fixed level. Since 27 July 2026, the law says so explicitly. | Tailor it per role. Legal works differently from marketing. | Regulation (EU) 2026/1744 |
| Form | Open. According to the Commission, just asking people to read the instructions for use might not work. | Hands-on workshops on people's own work, with a follow-up session a few weeks later. | European Commission, Q&A |
| Certificate | Not a requirement. You can keep an internal record of trainings and other initiatives. | A certificate of participation for each participant. That gives you part of your record right away. | European Commission, Q&A |
| Measuring knowledge | Not a requirement. | Know per role what people do with AI. | European Commission, Q&A |
| AI officer or fixed governance structure | Not a requirement. | Appoint one person who keeps track of the measures. | European Commission, Q&A |
The points on form, certificate, measuring and an AI officer come from the European Commission's guidance. That guidance is not legal text. The advice column is WAINUT's.
The law does not prescribe a form. According to the Commission, just asking people to read the instructions for use might not work. Training is therefore a logical measure: your people work faster and more safely with AI and you keep a record of what you did.
Then you tailor the measures to the role and context of your people.
The law leaves the form open. An e-learning is then quickly arranged. Our experience is that people work much the same after an e-learning alone. AI literacy grows when people do it themselves, on their own work, with a trainer who watches and corrects. Our trainings are built for that: hands-on workshops with your team's tasks. We always advise a follow-up session a few weeks later.
A practical approach in line with article 4 and the minimum the European Commission names.


Which AI systems does your organization use or develop, who works with them and what risk comes with them? That determines what your people need to know.
Everyone who works with AI on your behalf gets a general understanding of AI. What can it do and not do, where does it go wrong and how do you use it responsibly?
Take into account the knowledge, experience and training of your people and the context in which they use AI. Also consider the people on whom the AI is used, such as applicants or customers.
Agree which tools and which data are allowed and who people can turn to with questions. Then everyone knows what is expected.
Keep an overview of trainings, agreements and other measures. AI and the rules change, so update your approach when needed. The law itself changed on 27 July 2026.
According to the European Commission, a certificate is not needed. You can keep an internal record of trainings and other initiatives. Such an overview helps to show what you do.
Taking an AI literacy training with WAINUT? Then every participant receives a certificate of participation with a personal link. Anyone with the link or the QR code can check on academy.wainut.ai that WAINUT issued it. That gives you part of your overview right away.
Verify a certificate on academy.wainut.ai (opens in a new tab)
The AI Act applies in stages. These are the fixed dates. Since 27 July 2026, the dates for high-risk AI are fixed as well.

Article 4 and the prohibitions in article 5 apply. For example, AI that recognizes emotions at work or in education is prohibited, except for medical or safety reasons.
For models from before that date, the rules apply from 2 August 2027 at the latest.
The Digital Omnibus on AI enters into force. A fixed level of AI literacy for each person is no longer required. The dates for high-risk AI are fixed.
From then on the AI Act applies in general, including the transparency rules of article 50. According to the European Commission, supervision by national market surveillance authorities starts then. In the Netherlands, no supervisory authority has been designated yet.
AI for intimate images of an identifiable person without explicit consent and AI for child sexual abuse material are prohibited. Providers of existing generative systems must then also mark their output.
From this date, additional requirements may apply to AI in recruitment and selection, education, credit scoring or decisions on benefits. These include human oversight by people with the necessary competence, training and authority.
For products from annex I, section A, such as lifts, pressure equipment, personal protective equipment and ATEX equipment. This is mainly a duty of the manufacturer.
Final date for existing high-risk AI systems of public authorities.
In short: this has not been settled yet. This is the situation on 10 November 2026.
According to the European Commission, national market surveillance authorities supervise article 4. The draft Dutch implementation act works with ten market surveillance authorities. The AP and the Dutch Authority for Digital Infrastructure (RDI) coordinate. For article 4, the draft designates the AP, the AFM and DNB. It provides for an enforcement order (last onder bestuursdwang) and no administrative fine. This draft dates from before the Digital Omnibus and may still change.
Article 4 applies in every sector. The difference lies in the AI you use. Some applications may face additional requirements from 2 December 2027.
Public organizations have also had to take measures for the AI literacy of their people since February 2025. If you use AI to decide on benefits or other public services, additional requirements may apply from 2 December 2027, such as a fundamental rights impact assessment.
AI for admission, for assessing learning outcomes or for monitoring during tests may face additional requirements from 2 December 2027. AI that recognizes students' emotions has been prohibited since February 2025, except for medical or safety reasons.
Additional requirements may apply from 2 December 2027 if you use AI for credit scoring of individuals or for risk and pricing of life and health insurance. Fraud detection falls outside this.
Recruitment, assigning shifts based on behavior or performance and assessing employees: where AI plays a part in these, additional requirements may apply from 2 December 2027.
The question is whether AI serves as a safety component in the management or supply of gas, heating or electricity. If it does, additional requirements may apply from 2 December 2027.
Using AI to recruit staff or to manage roads or networks? Additional requirements may apply from 2 December 2027.
What WAINUT does
We train your people on the work they do every day. In the training and the adoption program, we map out what applies to you.


Don't see your question? Call 085 333 2077, email info@wainut.ai or book an intro call.
Taking measures is mandatory. Since 2 February 2025, article 4 of the AI Act has asked organizations that provide or use AI to take measures for the AI literacy of their people and of others who work with AI on their behalf. Since 27 July 2026, the law states that you do not have to ensure a specific level for each person. Which measures fit depends on your people and how you use AI.
The law does not prescribe a fixed form. Taking measures is required. According to the European Commission, just asking people to read the instructions for use might not work. Our experience: an e-learning alone changes little about how people work. Practicing on their own work with guidance does.
The law does not ask for a certificate. According to the European Commission, you can keep an internal record of trainings and other initiatives. After a WAINUT AI literacy training, every participant gets a certificate of participation that you can verify on academy.wainut.ai (opens in a new tab). That gives you part of your record right away.
According to the European Commission, measuring knowledge is not a requirement. Since 27 July 2026, the law also states that you do not have to ensure a specific level for each person. We do advise knowing per role what people do with AI. Then you tailor the training to it.
According to the European Commission, an AI officer or a fixed governance structure is not a requirement. We do advise appointing one person who keeps track of the AI literacy measures.
Yes. The AI Act also applies to public organizations. Since February 2025, every organization that uses AI has had to take measures for the AI literacy of its people. Since 27 July 2026, a fixed level for each person is not required. If you use AI to decide on benefits or other public services, additional requirements may apply from 2 December 2027, such as a fundamental rights impact assessment.
The AI Act sets no fine amount for article 4. Member states may set their own sanctions. In the Netherlands, that law is still being prepared. For article 4, the draft implementation act provides for an enforcement order (last onder bestuursdwang) and no administrative fine. That draft may still change.
That has not been settled yet. In the Netherlands, no supervisory authority for the AI Act has been designated by law. The Dutch Data Protection Authority is preparing to act as supervisor. The Dutch government plans to submit the implementation act to parliament in the second quarter of 2027.
This page was updated on . We update it when the law or its explanation changes.
Book a 30-minute intro call. Together we look at which measures suit your people and how you record what you did.
Or call 085 333 2077 or email info@wainut.ai