AI literacy and Article 4 of the EU AI Act

01What Article 4 requires

Article 4 does not prescribe one standard course or a universal certificate. It asks providers and deployers to adopt proportionate measures for the people who operate AI systems on their behalf. The obligation has applied since 2 February 2025; supervision and enforcement apply from 2 August 2026, the regulation's general application date.

Regulation (EU) 2026/1744 clarified that the duty does not require an organisation to guarantee that any individual reaches a specific level of literacy. That limits the outcome promised, not the duty to act: organisations still need appropriate measures, adapted to role, context and risk.

02What AI literacy means

Frameworks from bodies such as UNESCO and the OECD converge on a few competences: a basic grasp of how AI systems work, the ability to use them effectively, and the judgment to question their outputs and effects.

In practice: knowing that a model predicts rather than knows; recognising where automated systems shape your options; and asking who benefits, who is affected and who is accountable.

03Why regulation literacy belongs in AI literacy

A citizen who understands embeddings but not their rights in front of an automated rejection is only half-literate. The EU AI Act encodes society's current answers to the hard questions — which uses are unacceptable, which demand safeguards, when people must be informed.

Regulation literacy does not mean memorising articles. It means knowing that a risk-based framework exists, roughly how it sorts systems, and which questions it entitles you to ask.

04The digital citizenship connection

Rights and responsibilities around automated systems are civic matter: contestability of decisions, transparency about machine interaction, accountability of institutions that deploy AI. That is the territory of digital citizenship and AI regulation.

05Why scenario-based learning works here

Rules stick when they are needed, not when they are listed. Scenario-based learning puts the learner inside a situation — an opaque hiring filter, a scored neighbourhood — where the regulation's concepts become tools rather than trivia. The full argument is in serious games for AI regulation.

NO AI ACT applies this: in the game you classify 14 systems from evidence, and every concept — deployer, oversight, transparency — earns its place by being used.

06What a game can and cannot teach

• Can: the risk-based logic, the habit of evidence-grounded classification, the vocabulary in context, the sense that context of use changes everything.
• Cannot: the actual legal text, jurisprudence, compliance procedure, or anything requiring professional judgment. A simulation motivates and structures; it does not replace study — see the guide and glossary for the next layer.

07An evidence-based AI literacy classroom activity

For a short paired discussion, choose one game dossier and make four columns on paper: system use, affected people, available evidence and missing information. Work with the fictional case only, without adding real personal data.

1. Before classifying the system, identify an automated decision and someone who may be affected by it.
2. Compare two documents in the dossier: which supports your interpretation, and which challenges it?
3. Propose a measure and explain the specific problem it addresses. Your partner identifies missing information that could change the decision.

Finish with two sentences: “The decisive evidence is…” and “I would change my mind if…”. Use the educator session guide for facilitation and the risk-category examples for the classification discussion. This exercise does not certify competence or regulatory compliance.