Tailored training
AI literacy is now the employer’s duty – how to train your team right
· 5 min read
In short: Since 2 February 2025, the AI Act’s Article 4 has required every company that uses AI at work, whatever its size, to support its staff’s AI literacy. A sensible way to meet the obligation has three parts: shared basic training for all AI users, a role-specific module and written ground rules. Document the training, as the documentation shows when needed that the obligation has been met.
So the obligation also applies to an ordinary company that merely uses AI tools, not only to those who develop them. This article walks through what the law requires and how to meet the obligation sensibly.
What does AI literacy mean in the law?
AI literacy means the skills and understanding that allow a person to use AI systems with due consideration and to recognise their opportunities and risks. It is not about programming skills or deep technical expertise, but about a common-sense level of understanding: what the tool can do, where it goes wrong and what can safely be fed into it.
Under Article 4 of the AI Act, providers and deployers of AI systems must take measures to promote the AI literacy of their staff. The measures are proportionate to the staff’s technical knowledge, experience and training, and to the context in which AI is used. The requirement is therefore flexible: a customer service agent who uses a language model for draft replies needs different training from a developer who builds AI into a product.
Who does the obligation apply to?
In addition to providers, meaning companies that develop AI systems, the obligation applies to deployers, meaning organisations that use an AI system in their professional activities. If AI is used for work tasks in your company, for example for drafting texts, translating, producing images or supporting customer service, your company is a deployer and the obligation applies to it.
The obligation extends beyond your own staff to others who use AI systems on the company’s behalf, such as subcontractors. Company size does not matter, so a small company is covered as well. In practice, the sensible reading is this: if AI is part of the work, developing the skills to use it is part of the employer’s job.
What did the July 2026 amendment bring?
The EU amending regulation that entered into force in July 2026, the so-called digital omnibus, softened the wording of Article 4.
| Comparison | Original Article 4 (from 2 February 2025) | Amended Article 4 (July 2026) |
|---|---|---|
| Obligation | Ensure a sufficient level of AI literacy | Support the development of AI literacy |
| Level of competence | Not specified | The employer does not have to guarantee a specific level |
| Guidance | None | The Commission and member states will draw it up by 2 August 2027 |
In practice, the change does not remove the obligation but makes it more reasonable: what matters is that the company takes genuine, documented steps to build competence. What made sense before the change still makes sense, because a skilled team is in any case the best protection against mistakes in AI use.
Why is training worthwhile beyond the legal requirement?
The legal obligation is a good reason to start, but in practice the benefit is greater than ticking a regulatory box.
Most of the harm related to AI use comes from lack of skill, not from bad intent.
An employee trusts a figure a language model has invented and puts it in a customer quote, or feeds the customer register into a free tool whose terms nobody has read. Both situations can be avoided with a couple of hours of training.
The other side is the benefit that goes unclaimed. In an untrained organisation, AI is used only by the most enthusiastic, each in their own way, and good practices do not spread. Shared training brings the whole team to the same level and makes the use of the tools consistent, so the benefit also shows in the quality of the work.
What kind of training meets the obligation sensibly?
The good news is that the spirit of the law and the company’s own interest align here. A working package consists of three parts:
- Basic training for everyone who uses AI. What language models are, why they sometimes produce incorrect information, how results are checked and what data may be entered into the tools. It is especially important to agree that customers’ personal data and trade secrets are not put into tools whose data processing is uncertain.
- A role-specific deeper section. Marketing, customer service, finance and management use AI in different ways, so a short role-specific section makes the learning practical.
- Written ground rules. A one- or two-page guide on which tools are in use, what they are used for and what must not be entered into them. The guide keeps the knowledge alive as new employees join the company.
Training is not a one-off checkbox either. The tools evolve quickly, so a short refresher, say once a year and whenever new tools are introduced, keeps skills up to date with little effort.
Document the training sessions: who attended, when and what was covered. The documentation shows, if ever needed, that the company has taken the measures the law requires, and the same material serves as onboarding material going forward.
How do you get started?
If AI is already in everyday use in your company but training and ground rules are missing, the situation is easy to fix. Developit’s training services offer practical AI literacy training that is scaled to your team’s roles and tools and leaves your company with ready-made written ground rules. That way the obligation is taken care of and your team gets more out of AI, safely.
