Skip to main content

Is AI training mandatory? What Article 4 requires today

What the AI Omnibus changed, who the rule applies to and how to put it into practice and document it with little effort.

Updated: · 7 min read

Key points

  • There is no obligation to complete a specific training. Article 4 of the EU AI Act requires measures that support your team’s AI literacy, and training is the obvious way to do that.
  • This applies to every company where employees work with AI, even if it is only ChatGPT or Copilot. It has been supervised since 2 August 2026.
  • The regulation does not require a certificate, an AI officer or a specific duration. Internal records are sufficient as evidence.

A quote with a service your company does not even offer. A source that does not exist. A customer email that ended up in the prompt complete with the address. Recognising mistakes like these in your own daily work is what AI literacy is about. This tells you which instruction makes sense and how you can tell in daily work that it helps.

That is what Article 4 of the EU AI Act is about. It does not prescribe a specific course. It requires you to support your team’s AI literacy in a targeted way, matched to the tasks they use AI for.

Is there an obligation to train?

Not in the narrow sense. Article 4 does not name a mandatory training, a duration or a certificate. It requires measures that support the development of AI literacy. In practice, training is still the obvious route, because it is easy to plan and to document.

The rule applies to

  • every company that uses AI, regardless of size. Most are deployers: they use ready-made tools such as ChatGPT, Copilot or translation services.
  • companies that develop AI systems, as providers.
  • everyone who uses AI on behalf of the company. In its questions and answers on Article 4, the European Commission explicitly includes contractors and service providers.

What you need to organise

For Article 4, you organise five things:

  • Overview: who uses which AI tool for what, including accounts set up privately.
  • Roles: who uses AI rarely, who uses it every day, who selects tools or approves results.
  • Measures for each role: training and rules that fit these tasks.
  • Contact person: someone who answers questions and reviews new tools.
  • Documentation: a short document that records what you have put in place.

What your team should be able to do afterwards

The European Commission lists minimum points for an AI literacy programme: a general understanding of AI, your own role as provider or deployer, the risks of the system in question, and tailoring to prior knowledge and context of use. For teams that use ChatGPT for advertising copy or translations, for example, it becomes specific: they should be informed about risks such as hallucinations, meaning answers that sound convincing and are still wrong.

In daily work, this means that after a measure your team can

  • spot an invented source or figure in an AI answer and check it,
  • remove unnecessary customer and personal data from a prompt before sending it,
  • proofread an AI draft and correct mistakes before it leaves the company,
  • say which tools are allowed and whom to ask when unsure.

How you can tell it is working: mistakes in drafts are caught before they go out, and questions about data go to the contact person rather than into a private prompt. The Commission considers it ineffective in many cases to rely solely on an AI system’s instructions for use. That is why training should use real tasks from your daily work.

Which measures fit

Article 4 does not explicitly require training. Training is the obvious route. We recommend a combination of four building blocks:

Training by roleThe basics for everyone, more depth for those who work with AI every day.
Rules on useWhich tools are allowed and which data must not go into a prompt.
Contact personSomeone who answers questions and reviews new tools.
Regular refreshersAI tools change quickly. Refresh the knowledge when new tools and tasks come along.

Which measure suits whom depends on the task:

  • If someone uses AI rarely, for an occasional translation, for example, the basics and the rules are enough.
  • If someone uses AI every day to produce texts, quotes or analyses, add more depth: checking sources, reviewing drafts, protecting data.
  • If someone selects tools, approves results or leads a team, responsibility, contracts with providers and your own rules belong in the training too.
  • If a tool helps make decisions about people, for example on job applications or learning outcomes, the basics are not enough. Check whether it is high-risk AI.
  • If freelancers or service providers use AI on your behalf, agree in the contract which knowledge you expect and which rules apply.
Certificates of attendance are among the internal records that, according to the European Commission, are sufficient as evidence.

E-learning, training or both

The regulation does not prescribe a format. What matters is whether your team works more safely on its own tasks afterwards.

  • E-learning: good for the basics and for many people in different locations. For teams who work with AI every day, practice with their own tools and tasks is often missing.
  • Training or a workshop with the team: you practise on real tasks and clarify questions about your rules and tools straight away. On site or online.
  • Both: the basics by e-learning for everyone, plus training for those who work with AI every day.

Whatever the format, make sure there is a certificate of attendance with date, duration and content. Then it works as a record for your documentation.

The documentation, filled in

The AI Act does not prescribe any particular form. According to the European Commission’s questions and answers, internal records of training and other measures are sufficient. For each role, record the role, tool, typical task, risk, measure and the trigger for the next review. This is what it looks like for an engineering firm with 18 employees that uses ChatGPT, Copilot and a translation service with company accounts. Everyone has attended basic training, and a project manager is the contact person:

  • Quotes and sales (4 people): Tool: ChatGPT with a company account. Task: drafting quotes. Risk: invented services or dates in the draft. Measure: basic training with more depth on checking drafts, certificates of attendance filed. Next review: when a new tool is introduced, within a year at the latest.
  • Project management (5 people): Tool: Copilot in Microsoft 365. Task: summarising minutes and reports. Risk: clients’ project data in the prompt, wrong figures in summaries. Measure: training with more depth, rules on data. Next review: when new clients set their own requirements.
  • Administration (2 people): Tool: translation service with a company account. Task: translating emails. Risk: employee data in the prompt. Measure: basic training and rules on data. Next review: when new staff join.

Add the rules with their date, the contact person and the certificates of attendance. For a small team, this fits on two pages. Article 4 does not require a dedicated AI officer. The Commission makes clear that no particular organisational structure is prescribed.

Background: what the AI Omnibus changed

The AI Omnibus, Regulation (EU) 2026/1744, amended Article 4 with effect from 27 July 2026. Previously, providers and deployers were to ensure, to their best extent, that their staff had a sufficient level of AI literacy. Now Article 4 requires measures that support the development of AI literacy. Nobody has to guarantee a specific level for individual people any more. Technical knowledge, experience, education and training, the context of use and the people on whom AI is used are still taken into account.

Who checks and from when

  • 2 February 2025: Article 4 starts to apply.
  • 27 July 2026: The AI Omnibus amends Article 4 to its current wording.
  • 2 August 2026: The national market surveillance authorities start supervising compliance.

According to the European Commission’s questions and answers, the authorities can also impose penalties for breaches of Article 4. Any penalty must be proportionate in the individual case. The Commission considers penalties more likely where an incident can be traced back to a lack of training and guidance. All dates under the AI Act are on our topic page EU AI Act.

Next steps

These steps put Article 4 into practice in your organisation:

  • Ask your team which AI tools they use, including ones they have set up privately.
  • Assign people to the roles and define for each role what they should be able to do afterwards.
  • Define approved tools and rules, see Can my team use ChatGPT?
  • Plan training for each role and name a contact person.
  • Set up the documentation as in the example above and enter the trigger for the next review.

If you would rather not set up the training yourself, our AI training under the EU AI Act shows how your team checks AI results and protects confidential information, and includes a certificate of attendance for each person.

Worksheet as PDF: Checklist: AI literacy under Article 4

The templates and checklists from this article to print and fill in. You get the PDF when you subscribe to the lindenbrook newsletter with new articles and worksheets. The newsletter is in German.

By clicking, you subscribe to the newsletter. You will receive an email straight away. The link in it confirms your subscription and takes you to the PDF. You can unsubscribe via the link in every issue. The newsletter is sent via bemailed, which measures opens and clicks. More in our privacy policy.

Frequently asked questions

Does the obligation also apply to small companies?

Yes. Article 4 makes no distinction by company size. The type and scope of the measures can, however, reflect how AI is actually used. The Commission and the Member States are to support small and medium-sized enterprises in particular.

Is e-learning enough for everyone?

E-learning can be one building block, especially for the basics. The measures should fit the tasks: someone who writes texts for customers with AI every day needs more than someone who uses it rarely. The European Commission considers relying solely on instructions for use ineffective in many cases.

How much does AI training cost?

That depends on the format. E-learning is usually charged per person, training and workshops for a team per session. Free courses, for example on KI-Campus, can be one building block for the basics. For Article 4, what counts is that the measures fit your team’s tasks.

Do we need a certificate or an AI officer?

No. According to the European Commission’s questions and answers, no certificate is needed. Companies can keep internal records of training and other measures. Article 4 does not prescribe any particular organisational structure, such as an AI officer.

Is an AI policy mandatory?

Article 4 does not require an AI policy. Written rules on use are, however, one of the measures that support AI literacy, and they make your measures traceable. A template with everything that belongs in it is in the guide “AI policy: a template to adapt”.

How often should training take place?

The AI Act does not set an interval. When there are new tools, new tasks or new employees, plan a refresher.

Sources

  1. EU AI Act, Regulation (EU) 2024/1689, consolidated version of 27 July 2026, EUR-Lex
  2. Regulation (EU) 2026/1744 (AI Omnibus), EUR-Lex
  3. European Commission: AI Literacy, Questions and Answers

Updated: . This article is not legal advice.

AI training under the EU AI ActHalf a day · on site or online · certificate of attendance for each participant
See the training

Read next

Questions about your situation?

Write to us. We will get back to you personally with a proposal for the next steps.