Skip to content
AImori
EN
Free for Chrome
Menu

AI literacy under AI Act Art. 4: what staff training should cover

Since 2 February 2025, Article 4 of the EU AI Act requires companies to support the AI literacy of their staff. This does not mean a computer science degree but everyday knowledge: what an AI service can do, where it gets things wrong and which data must never go into it. This guide explains what the duty means today and what a short, effective training looks like.

What Article 4 requires

  • Who: providers and deployers of AI systems. A company that uses AI services such as ChatGPT, Copilot or Gemini professionally is a deployer.
  • What: since the new wording by amending Regulation (EU) 2026/1744 ("Digital Omnibus", in force since 27 July 2026), deployers must take measures that support the AI literacy of their staff. No specific level is required.
  • What AI literacy means: skills, knowledge and understanding to deploy AI systems in an informed way and to be aware of the opportunities, risks and possible harm (Art. 3 point 56).
  • Format: the regulation prescribes no course, duration or certificate. What matters is that the measures fit the company, the AI services it uses and the tasks of its staff.

Does it apply to Swiss companies?

Not directly. The AI Act affects Swiss companies if they are established in the EU, offer AI systems in the EU or if the AI output is used in the EU (Art. 2). Regardless of that, the Swiss Data Protection Act applies directly to AI applications, as the Federal Data Protection and Information Commissioner (FDPIC) points out. Training helps either way: it lowers the risk that personal data or trade secrets end up in an AI chat.

What AI training should cover

One hour is enough to start, as long as it is concrete. These six topics work well:

  1. Which services are allowed: with which account (company, not personal) and for what.
  2. What never goes in: customer data, bank details, national ID numbers, passwords, health data, confidential contracts. More in the guide IBANs, ID numbers & co..
  3. How AI gets things wrong: answers sound confident but can be wrong or made up. Always check facts, figures and sources.
  4. Humans stay responsible: read and approve texts, code and decisions from AI before passing them on.
  5. Copyright and confidentiality: do not upload other people's documents without permission, do not publish results unchecked.
  6. What to do after a mistake: whom to tell and how quickly.

Keep a record of who was trained and when. The free AI policy template has a dedicated "Training" field for this.

Making training work in everyday life

Knowledge alone does not prevent mistakes under pressure, for example when pasting a whole email thread. Two aids anchor the training:

  • Practise with real examples: let employees check one of their own texts in the free test "Is it safe to paste into ChatGPT?". The test runs only in the browser, nothing is sent.
  • A reminder at the right moment: AImori checks input on ChatGPT, Claude, Gemini and Copilot locally in the browser and shows what it found before sending. The rule from the training becomes a hint exactly when it is needed. AImori is an aid against mistakes and replaces neither training nor policy.

There is also an AI literacy record: every warning from AImori explains in one sentence why the value is sensitive. From the explanations counted locally, a record can be printed – numbers per data type group and month only, no content – that companies can use for their measures under Art. 4. It is a self-declaration, does not prove that obligations are met and does not replace training.

AImori is free. Companies deploy the extension via Intune or Group Policy and set the rules centrally. Get AImori · For companies

Read next: Shadow AI in small businesses · Data loss prevention (DLP) for AI tools

Frequently asked questions

Do we need a certificate for AI training?

No. Article 4 prescribes no specific format. A short internal training with examples from your own work and a written note of who took part is a good start.

How often should we train?

When people join and regularly after that, for example once a year or when new AI services are added. The regulation sets no fixed interval.

Does Article 4 apply if we only use ChatGPT?

Yes, provided the AI Act applies to your company (see above): anyone who uses an AI service professionally is a deployer. Whether this applies to your company is a question for a specialist in the individual case.

Sources

Information as of 17 October 2026. This guide is not legal advice.

Free for Chrome