Is there an obligation to provide AI training?
Yes, even though the law doesn't use the word "training obligation". Article 4 of the EU AI Act (Regulation (EU) 2024/1689) requires companies that provide or professionally use AI systems to take measures on their staff's AI literacy. The obligation has applied since 2 February 2025.
AI literacy means: anyone who works with AI understands what the tool can do, where it makes mistakes and what risks can arise. The law doesn't prescribe how you get there. Training is, however, the most obvious way and the easiest to document.
Who is affected?
The obligation applies to providers and deployers of AI systems. A deployer is any company that uses an AI system professionally under its own authority. So it doesn't only concern software vendors, but almost every business where employees use ChatGPT, Microsoft Copilot or an AI translation tool for their work, for example.
There is no exemption based on company size. Besides your own employees, it also covers people who work with the AI systems on your behalf, such as freelancers or external service providers.
What did the 2026 AI Omnibus change?

With the so-called AI Omnibus (Regulation (EU) 2026/1744), the EU adjusted several rules of the AI Act in summer 2026, including Article 4. The obligation was not removed but recalibrated:
| Before the AI Omnibus | After the AI Omnibus | |
|---|---|---|
| What is required | Measures to ensure, to the best extent, a sufficient level of AI literacy | Measures that support the development of AI literacy |
| Outcome | A sufficient level should be reached | No specific level has to be guaranteed |
| Support | No explicit role for authorities | The Commission and Member States are to support companies with examples and recommendations |
In practice, this means you don't have to prove that every person has reached a certain level of knowledge. But you do have to be able to show that you have done something.
Do you need a certificate?
No. In its questions and answers on AI literacy, the European Commission makes clear that no external certification is required. It recommends keeping an internal record of the measures taken. A certificate can still be useful, for example as simple proof for customers or in tenders.
What happens if I do nothing?
The AI Act does not set a specific fine for breaches of Article 4. Supervision lies with national authorities, and the enforcement rules have applied since August 2026. Another point matters more: if damage occurs because employees used an AI tool incorrectly, a lack of training can count against your company. This article does not replace legal advice.
AI literacy in your company in 5 steps
- Inventory: which AI tools are used in the company, officially and unofficially? Who uses them for what?
- Form groups: not everyone needs the same. Someone who occasionally drafts texts needs less than someone who analyzes customer data with AI or looks after an AI agent.
- Train: basic training for everyone, plus deeper sessions by role. Practical, with the tools that are actually in use.
- Set rules: a short internal AI policy defines which tools are allowed, which data may go in and when results must be checked.
- Document and refresh: record who took part in what and when. New tools or applications mean a refresher.
What belongs in AI training?
There is no fixed curriculum. The content should fit prior knowledge, role and the systems in use. These building blocks have proven useful:
- Basics: how do language models work, and why do their answers sound convincing even when they are wrong?
- Limits and risks: made-up facts, outdated knowledge, bias, dependence on the input.
- Data protection and confidentiality: which data may go into which tool. More in the article on AI and data protection.
- Good prompts: describe tasks clearly, give context, check results and refine.
- Your rules: the internal AI policy, approvals and contacts.
- Practice: exercises with real tasks from the group's daily work.
An example: a business with 40 employees

Imagine a mid-sized business where sales and administration use an AI assistant for texts, accounting is testing AI receipt recognition and two people look after a chatbot on the website. Implementation could look like this:
- All 40 employees get two to three hours of basic training on how AI works, its limits, data protection and the new AI policy.
- Sales and administration go deeper in a half-day workshop on the text assistant, using their own tasks as exercises.
- Accounting learns how to check the results of receipt recognition and report errors.
- The chatbot team also covers transparency obligations, the bot's limits and how to handle complaints.
- Documentation: a simple list with name, date, content and duration, plus the training materials and the policy.
With manageable effort, this creates a record that shows the company is actively supporting AI literacy.
Common mistakes
- Training only managers. The obligation covers everyone who works with AI.
- One training for all. A general talk helps little if the tools and tasks in everyday work are completely different.
- Overlooking shadow use. Private AI accounts used for work belong in the inventory.
- Documenting nothing. Without attendance lists and content, nothing can be proven later.
- Once and never again. AI tools change quickly. New features need a short update.
Checklist: AI literacy under Art. 4
- All AI tools used in the company are recorded, including unofficial ones.
- It is clear what knowledge each group needs.
- Basic training for everyone has taken place.
- Roles with more AI responsibility have had deeper training.
- An internal AI policy has been adopted and communicated.
- External service providers working with AI on your behalf are included.
- Attendance, content and dates of the training are documented.
- A date for the next refresher is set.
How uberneu supports you
We train teams in a practical way with the tools you actually use, from basic training to workshops for individual departments. On request, we help with the inventory, the AI policy and documentation, and if you then want to integrate AI more deeply into your processes, we also support the AI integration.




