AI Literacy und KI-Kompetenz: Definition, Stufen und Artikel 4

AI literacy is the ability to use AI systems in your own work context in an informed, critical and responsible way. It includes understanding what a system can and cannot do, judging its output, and weighing legal and ethical consequences. Under the EU AI Act, this ability is now a duty for companies.

Competence that lands in the process. In just one week your team builds on real cases and documents its learning. Book a discovery call, or enter your email below.

Literacy first meant reading and writing. Over decades it produced derived competence terms such as media literacy and data literacy, each the basic ability to deal with a medium responsibly. AI literacy stands in that line. The shift came in 2025. Before, it was a soft educational goal, nice to have but non-binding. Since Article 4 of the EU AI Act took effect on 2 February 2025, it is a requirement for providers and deployers of AI systems. The term changes register, from a description of skill to a duty with an addressee and a context. That is why AI literacy is worth treating not as a buzzword but as something you must build and document in a demonstrable way.

AI literacy is not a yes or no quantity, it is a ladder. Reaching only the bottom rung means operating a tool without being able to judge it. The higher rungs carry responsibility. In practice it helps to name the levels and map roles onto them.

  Level 4  Governance   set rules, assess risk, own the deployment
  Level 3  Judge        check output, see limits, spot hallucination
  Level 2  Apply        solve tasks with AI, steer prompts
  Level 1  Operate      open a tool and run a basic task
     |
     +--> Article 4 asks for the level that fits each role, not the top level for everyone

The key is the arrow at the bottom. The EU AI Act does not ask everyone to reach the governance level. It asks for competence that fits the role and the risk. A caseworker using an approved tool needs level 2 and a feel for level 3. Whoever decides on AI deployment needs level 4. That role logic is the difference between a meaningful competence measure and a tick-box training for all.

An illustrative model, not a client figure. A 200-person company wants to take Article 4 seriously. Instead of rolling a 60-minute mandatory course out to everyone, it maps roles to levels.

Effort concentrates on the 48 people where competence really carries impact and risk. The numbers are a planning model, not a rule. The point is the structure, since one course for all is expensive and hits the wrong people, while a role-based mapping is cheaper and more defensible.

The pressure is the same everywhere, the consequence is not. In finance, insurance and healthcare the cost of a misjudged AI output is highest, so the focus is levels 3 and 4 and documentation. In marketing agencies, our first target market, the need is broad application skill and clean checking of sources and rights. In HR and recruiting the discrimination risk leads, invisible without competence. In industry and IT competence meets security, because data and systems are touched. The common thread is that competence is measured against the role, not against an attendance certificate.

Literacy is the ability itself. Everything else is the path to it, the frame around it, or a record of it.

It is urgent when AI is already in use, when staff use AI in customer-facing or data-facing processes, and when Article 4 documentation is missing. It can wait when AI is not used and not planned, because then the application context that competence attaches to is absent. Competence without application fades, so the best moment is when real use begins.

Since 2 February 2025, Article 4 of Regulation 2024/1689 requires providers and deployers to ensure a sufficient level of AI literacy among their staff and others operating on their behalf. Sufficient means role and context specific, measured against technical knowledge, experience, education and the purpose of use. A practical competence measure, such as building on real cases with a documented learning state, can support and evidence this duty. It is not an official certificate and does not guarantee automatic conformity. The company assesses adequacy itself, with qualified counsel where needed.

Article 4 scope note: this page explains the competence concept and how a measure can support and document competence building. It is not legal advice and not proof of automatic conformity.

AI enablement · AI upskilling · AI adoption · AI certification · AI skills gap matrix

Two ways, depending on where you are.