Article 4 changed: your people need a record, not a score

Something changed on 27 July 2026 that most AI literacy programmes have not caught up with yet. Regulation (EU) 2026/1744 rewrote Article 4 of the EU AI Act. The duty is still there, still direct, still on providers and deployers. But its shape moved.

The old text put the outcome on you: your people had to reach a level that counted as adequate. The new text asks you to take measures that support the development of AI literacy, and says explicitly that you do not have to guarantee a specific individual level.

Read quickly, that sounds like less work. Read properly, it moves the burden from your people to your records.

Why a lighter duty is harder to prove

Under the old wording you could, in theory, point at a result. Someone passed the assessment, so the level was there. It was a thin argument, but it was an argument.

Under the new wording that route is closed. The duty is not about what one person knows. It is about what you did as an organisation, and whether that was appropriate to the technical knowledge, experience, education and training of the people involved, to the context in which the systems are used, and to the people those systems are used on.

There is no test that ticks that box. There is only the record of what you decided, for whom, and why.

What that record has to contain

Four things, in this order. Skip the first and the rest becomes generic, and generic is exactly what the text does not ask for.

Which people work with which systems. Not headcount. Names and roles mapped to the AI systems they actually touch. A recruiter using AI in shortlisting has a different exposure than a marketer generating copy, and the measure has to reflect that.

What each role needs to recognise. Write it as behaviour, not as a course title. "Recruiters can explain when AI use has to be disclosed to a candidate" is assessable. "Recruiters completed AI awareness" is not.

What you delivered and when. Learning path, scenario practice, assessment, date, result. This is the part most organisations already have somewhere, usually scattered across an LMS export and a spreadsheet.

What you did about the gaps. This is the part almost nobody has, and it is the part that turns a list into evidence. A group that fell behind and stayed behind shows that you offered something. A group that fell behind and was followed up shows that you took a measure.