Text type: EDT (Editorial/Classification) | Factual status: 18 August 2026
Since early August, further provisions of the AI Act have come under enforcement by the European Commission and national supervisory authorities. This directs attention to a provision that might initially seem inconspicuous: Article 4. In its current version, it requires providers and deployers of AI systems to take measures to foster AI literacy among the individuals who operate or utilize these systems on their behalf. The Commission clarifies: This entails neither a mandatory individual proficiency score nor a certificate for each individual person.
This is an essential clarification. AI literacy is neither a driving license one passes once and for all, nor an excuse to delegate responsibility to a training slide. The standard is context-dependent: who uses a system for what purpose, what prior experience exists, and which groups of people might be affected by its deployment. These are precisely the factors referenced in the current Article 4.
No Certificate — But a Duty to Appropriate Practice
The obligation is less convenient than a standardized certificate. A training course can certainly be sensible, yet it does not automatically answer the decisive questions: Are system outputs scrutinized? Do staff members understand the operational boundaries of the tool? Can affected parties identify when AI is involved? And is there a designated person or entity responsible for rectifying mistakes?
The European Commission therefore defines AI literacy as a duty for providers and deployers. Technical knowledge, experience, education, and training must be taken into account—alongside the context of use and the individuals or groups upon whom a system may impact. Since 2 August 2026, enforcement of additional AI Act provisions has commenced; the Commission explicitly highlights the new transparency mandates for specific systems and content. The corresponding announcement of 31 July makes this temporal nexus clear.
This does not mean that every organization requires the exact same training architecture. An internal utility for drafting texts raises different questions than a system that prescreens job applicants, assesses benefits, or prepares public administrative decisions. Yet in none of these instances is it sufficient to merely gaze at the user interface. Competence is demonstrated by whether individuals can identify the risks and responsibilities inherent in their specific operational work.
Four Competencies Instead of a Badge
For everyday practice, this translates into four core competencies:
First: Source Competence. A convincingly phrased AI-generated text is not evidence. Anyone who adopts generated content must be able to discern when an assertion demands a verifiable source, how that source can be substantiated, and when uncertainties must be made explicit. This applies with particular urgency where legal outcomes, public health, administrative decisions, or public discourse are involved.
Second: Role Competence. A model can assist with sorting, translating, or phrasing. It cannot, however, assume the accountability of an editorial office, a specialist division, or a public authority. Roles must therefore remain clearly visible: Who drafted the text, who reviewed it, who made the decision, and who is reachable for inquiries or corrections?
Third: Boundary Competence. Not every task ought to be automated. With personal data, complex legal questions, clinical information, time-sensitive filings, or high-impact public statements, there is a recurring need for supplementary human review, detailed documentation, or an intentional halt. Sound practice defines not only authorized tools, but also the concrete boundaries where the tool ceases to be sufficient.
Fourth: Rectification Competence. An absence of errors is not a realistic promise. The decisive criterion is whether errors are noticed, reported, and traceably corrected. This ability links AI literacy directly with transparency: Those who disclose responsibilities and review channels make criticism constructive in the first place.
What Educational Frameworks Make of This
The debate is not confined to enterprises. The UNESCO AI Competency Framework for Teachers outlines 15 competencies across five dimensions: human-centred mindset, ethics of AI, foundations and applications, AI pedagogy, and professional development. It arranges these not as a single examination, but across three developmental stages: acquire, deepen, and facilitate.
This is not a direct statutory interpretation of the AI Act, nor a blueprint that organizations must adopt verbatim. However, it illustrates a vital principle: competencies mature through operational practice and experience. The responsible adoption of AI therefore requires more than an introductory seminar; it demands opportunities to practice on concrete cases, discuss discrepancies, and refine operational guidelines.
For Newsrooms: Transparency Is an Operating Principle
For newsrooms, this question becomes especially tangible. Those who utilize AI for research, structuring, or phrasing need not sensationalize every minor note. But for published editorial content, it must be comprehensible who bears responsibility, how sources were substantiated, and how readers can submit corrections. The transparency rules of the AI Act do not replace this journalistic diligence; they simply make the imperative of accountability more visible.
A disclosure tag can indicate that AI was involved. It does not answer whether an assertion is accurate, whether someone substantiated it, or whether a correction mechanism exists. That is why AI literacy is not a badge or sticker. It is an operating principle: verify assertions, separate roles, recognize boundaries, and enable corrections.
That is less spectacular than debating whether a single graphic displays a watermark. But for a public sphere that must rely on accurate information, it is the far sounder foundation.
Sources
- European Commission: AI talent, skills and literacy – Overview of Article 4 and the Digital Omnibus.
- European Commission: AI Literacy – Questions & Answers – Application of Article 4, documentation, and lack of individual certification requirement.
- European Commission: Commission starts enforcing AI Act rules and new transparency requirements on 2 August – Communication of 31 July 2026.
- EUR-Lex: Regulation (EU) 2026/1744, Article 4 – Current wording of the AI literacy requirement following the Digital Omnibus.
- UNESCO: AI competency framework for teachers – Global educational framework, 2024, updated 2026.
Transparency and Responsibility: This article was drafted with assistance from GPT-5.6. Prior to publication, all sources were verified against primary documentation; editorial responsibility and final text verification remain with Um:bruch. Legally responsible under German press law is Lukas Geiger (V.i.S.d.P. pursuant to § 18 para. 2 MStV).
Translation: Gemini (GM). Review: Claude (CL). In case of discrepancies, the German version prevails.