In many companies, Article 4 of the EU AI Act produced one thing: an e-learning module everyone clicks through once, and an attendance list in a folder. Few people came out better at using ChatGPT or Copilot safely in their own work.
Since July 2026 there is new confusion: the Digital Omnibus rewrote Article 4, and some read it as the end of the duty. It is not. Below: what changed, what to record, and which format fits, with sources.
AI literacy training under EU AI Act Article 4
AI literacy means the skills and understanding to use AI in an informed way, aware of its opportunities and risks. Article 4 (Regulation (EU) 2024/1689) has applied since February 2, 2025 to providers and deployers of AI systems. A company that gives its staff ChatGPT, Microsoft Copilot, or another AI tool is a deployer, whatever its size or sector. The Digital Omnibus on AI (Regulation (EU) 2026/1744, in force since July 27, 2026) rewrote the article:
| Version from February 2025 | Version since July 27, 2026 | |
|---|---|---|
| Core wording | Take measures to "ensure, to their best extent, a sufficient level of AI literacy" | Take measures "to support the development of AI literacy" |
| Type of duty | Effort toward a sufficient level | Duty to take measures; expressly no guaranteed level for any individual |
| Who is covered | Providers and deployers, for staff and others acting on their behalf | Same |
| To take into account | Knowledge, experience, education, context of use, affected people | Same |
| Support | Not addressed | Commission and Member States support companies, in particular SMEs |
| Format, certificate | Not prescribed | Not prescribed |
What stays the same: the obligation. The European Commission writes in its Q&A that AI literacy remains an obligation for providers and deployers; only a specific "sufficient" level is no longer mandated. Pointing staff to the instructions for use is not enough on its own, and staff using ChatGPT should be told about risks such as made-up facts. Deployers of high-risk AI must still train staff for human oversight.
For US and UK companies with EU operations
Under Article 2, the Act applies to deployers established or located in the EU, wherever headquarters sit, so EU offices and subsidiaries that use AI tools are in scope. It also reaches providers and deployers outside the EU when the output of their AI system is used in the EU. For most US and UK groups, the practical answer is one program for the EU teams that also serves everyone else.
What to record
The Commission asks only for an internal record. Two national regulators spell out more:
| Item | RTR AI Service Desk (Austria) | Bundesnetzagentur (Germany) |
|---|---|---|
| Type of measure | Type of training (in person, e-learning) | Type of measures |
| Organizer | For external training | Not listed |
| Who | Per person concerned, in the personnel file | Participants |
| Content | Content and goal of the training | Scope of content |
| Date | When the training took place | Duration |
| Repetition | Repetitions | Regular evaluation, refreshers as needed |
| Framework | Written AI strategy, AI policy, training concept | Set goals and responsibilities |
Not required, according to the Bundesnetzagentur: guaranteed literacy levels, standardized training, external certification, or an AI officer.
Which format fits
The Act prescribes no format; what counts is fit with role, tool, and risk. Specialty Tokens wrote this page; the last row is our format.
| Format | Fits when | Doesn't fit when |
|---|---|---|
| E-learning or online AI literacy course | Many people need the basics and new hires join all the time | Teams need to use a tool well on real work |
| Free resources (the Commission's living repository of practices, European Digital Innovation Hubs) | A small company is taking first steps | Your own context of use needs training |
| Internal training | You have in-house AI experience and time | Nobody knows the approved tools in depth |
| In-person training by role, on real tasks | A tool is approved and teams should use it productively | A short awareness briefing is all you need |
For a side-by-side of providers, us included, see corporate AI training providers compared.
How we train
- Leadership: where AI pays off, what the risks are, who decides. See AI training for executives. As a partner of the Austrian Federal Economic Chamber (WKO), we run a series of in-person executive AI sessions.
- Business teams: the approved tool on their own tasks, in ChatGPT training, Microsoft Copilot training, or Claude training.
- Power users: people who build workflows for their department, up to the AI Champions Program.
- Engineers and IT: agentic coding, system integration, and security.
Typical content across roles:
- How AI models work: why they sound convincing and still make facts up.
- The approved tools on your reports, inboxes, and analyses, connected to your systems.
- Data rules: personal data and confidential information.
- Checking output: asking for sources, reviewing, staying accountable.
- Your AI policy, explained on concrete cases rather than read aloud.
Literacy shows in the work: at TimeTac, a marketing and growth team with no engineers had 10 automations live after an AI enablement workshop and an internal AI platform.
After each session we record format, organizer, content and goal, participants, date, and the planned refresher. Sessions run on site in the EU, the UK, and the US.
This page is not legal advice; your legal team should confirm what applies to you. The training is part of our AI training for employees. To plan which roles need what, get in touch.
