Skip to content
Beyond Prompt AI Studio
The EU AI Act in Practice

The AI Literacy Duty: enforceable from 2 August 2026

While everyone stares at the big high-risk deadlines, one duty has long been in force and applies to literally every company using AI: the AI literacy duty. And its window is tightening fast – national enforcement begins on 2 August 2026.

Four practical cases – worth remembering

Try it yourself: match the role to its training depth

Role

Training depth

What Art. 4 literally requires

Providers and deployers of AI systems must ensure that their staff and everyone else who handles AI systems on their behalf have a sufficient level of AI literacy. The key word is “ensure”: it's not enough for competence to somehow exist – the company has to actively make it happen, and be able to demonstrate that if questioned.

Since when does it apply, and when does enforcement start?

Art. 4 has already been applicable law since 2 February 2025 – unusually early compared to most other AI Act obligations. What's been missing so far is enforcement pressure: national market surveillance authorities (in Germany, the Bundesnetzagentur, via a new coordination and competence centre) only start actively checking this duty on 2 August 2026. For companies that means: the duty itself isn't new, but the real risk of being asked about it is rising sharply right now.

What counts as evidence – and what doesn't

The legislator deliberately prescribes neither content nor format – that creates flexibility, but also uncertainty. In practice, the duty usually boils down to a structured training measure. What doesn't cut it: handing out a PDF user manual without comment, or releasing a tool with no framing at all. What counts as a solid foundation in practice: documented training with recognisable content, proof of attendance, and a date – tiered by role, not a one-size-fits-all format.

Practice section: an evidence building block in three steps

First, tie training depth to the role: someone who only consumes a finished tool needs different content than someone building their own prompts, workflows, or models. Second, document the process – who attended, when, with what content, written down rather than just assumed. Third, repeat it whenever new tools are introduced or risks change – literacy isn't a one-time checkbox, it's an ongoing process. A completed, certified course like this one can serve as one building block of such evidence – but it doesn't replace a company-specific review or case-by-case legal advice.

The key points

  • Art. 4 has applied since 2 February 2025 – but enforcement by national market surveillance authorities only starts on 2 August 2026.
  • In Germany, the Bundesnetzagentur bundles this responsibility via a new coordination and competence centre.
  • Neither content nor format is legally prescribed – that creates flexibility but requires your own documented concept.
  • Training depth should follow the role: consumers, prompt/workflow builders, and model developers need different depths of competence.
  • A completed, certified course can serve as one building block of evidence – but doesn't replace a company-specific review or case-by-case legal advice.

The EU AI Act: what companies really need to know – and what's just panic

Quick check: did it sink in?

1 / 3

Since when has the AI literacy duty (Art. 4) legally applied?

Want to set up documented AI literacy training for your team?