Half your company uses ChatGPT or Copilot every day. Some of them on a company account, some on their own phone. Then a client's procurement form lands on your desk with a new line: "Describe the AI training provided to your staff." And you remember reading somewhere that the EU softened the AI literacy rule this summer. So, can you skip it?
The short answer, as the law stands on 15 October 2026: no. Article 4 of the EU AI Act still requires every company that uses AI at work to take measures, in practice training, for the people who use it. The July 2026 Omnibus changed the wording, not the duty. You no longer have to guarantee a "sufficient level" for each person. You do have to take measures that fit how your people use AI, and be able to show them. No certificate is required. An internal record is enough.
What Article 4 says since the July 2026 Omnibus
The original text of Regulation (EU) 2024/1689, in application since 2 February 2025, asked companies to ensure, to their best extent, a sufficient level of AI literacy among their staff. Nobody could tell you what "sufficient" meant. That word is gone.
Regulation (EU) 2026/1744 of 8 July 2026, the AI Omnibus, replaced Article 4. In plain words, the new version says:
- providers and deployers of AI systems take measures to support the AI literacy of their staff and of anyone operating or using AI on their behalf;
- those measures take into account people's technical knowledge, their experience, their training, the context the AI is used in and the people it is used on;
- no specific level of AI literacy has to be guaranteed for any individual;
- the Commission and the Member States support companies, small and mid-sized ones in particular, and the Commission publishes practical examples of compliance.
Here is the part most summaries skip. When the Commission presented its Digital Omnibus on 19 November 2025, it proposed to move the duty away from companies and onto the Member States and the Commission. That is what its own AI literacy Q&A describes. The final text did not do that. The obligation stayed with the companies that use AI. If you read "the EU dropped AI literacy" in July, you read about the proposal, not the law.
Who it covers
If your company uses an AI system at work in the EU, you are a "deployer" in the language of the Act, and Article 4 applies to you. There is no exemption for size or for industry. The Act also reaches companies outside the EU when the output of their AI is used inside it (Article 2).
"Staff and other persons" is broader than your payroll. A freelancer who writes your newsletter with an AI tool, or the agency that runs your ad campaigns with generative tools: both operate AI on your behalf.
And no, you don't need an AI officer. The Commission's Q&A says that no specific governance structure is mandated for Article 4. Someone in the leadership team still has to own it. In practice, if nobody is named, nobody does it.
What counts as a measure
This is the question I get. Not "which course should we buy", but "what will a regulator, or a client, accept as proof". The text gives you the criteria: the people, and the context they use AI in. A one-hour video sent to every employee doesn't meet them. It treats the recruiter who screens CVs with an AI tool the same way as the assistant who summarises meetings, and the law asks you to tell them apart.
The method we use with clients fits in four moves:
- map who uses which AI tools, including the ones nobody approved (that's your shadow AI);
- set the expected level per population: the leadership team that decides the AI strategy, the teams that use assistants every day, the people who oversee a sensitive tool;
- train each group on its real use cases, starting with the leadership team;
- write it down: the plan, the content, who attended, when.
The last move is the one that counts on the day someone asks. The Commission's Q&A (consulted on 15 October 2026) is explicit: there is no need for a certificate, and an internal record of trainings is enough. A spreadsheet works.
Why the leadership team first? Because the questions that block adoption are leadership questions. In our bootcamp sessions in October 2026, the leaders didn't ask about prompts. They asked which accounts their teams were allowed to use, and how to bring employee representatives into the AI charter before rolling anything out. Those answers have to exist before you train everyone, or the training will contradict the policy.
One case we described in our EU AI Act article for executives: a CEO ran an AI literacy assessment across his 120 employees, and 14% reached a baseline level. He started a six-week training plan. That baseline, by the way, is already a measure you can show.
What happens if you do nothing
Supervision of Article 4 sits with national market surveillance authorities, not with the EU AI Office, and the Commission says they supervise and enforce it from August 2026. The Act sets no specific fine amount for Article 4. Penalty rules are left to each Member State (Article 99), so the exposure depends on the country.
Honestly, I see the commercial risk arrive faster than the regulator. Large clients now ask suppliers to describe their AI governance and training. A company with nothing written down loses points in a tender long before it meets an inspector. That is what we see with our clients, not a statistic.
One rule did not move at all. If you deploy a high-risk system, for instance a tool that filters job applications, the people who oversee it must be trained and have the authority to overrule it (Article 26). Those obligations start on 2 December 2027. We covered that case in AI in hiring: what the EU AI Act changes.
The plan for this quarter
Name one owner in the leadership team this week. Run the inventory at your next leadership meeting: each director lists the AI tools in their area, approved or not. Train the leadership team first, then each population on its own use cases. Keep the record from day one, not after the last session.
If your teams already use AI and you have never trained anyone, you are not late by months. You are late by one decision. For the team side, read training teams on AI without the pushback.
Our AI training for executives and team adoption programmes are delivered in French today, with the training records your compliance file needs (the French page is here). If your team works in English, join the waitlist and you will hear first when an English programme opens.
Join the waitlist →
