EU AI Act in practice9 min read

EU AI Act Article 4: what your staff need to know about AI literacy

Contents

EU AI Act Article 4 does not prescribe a training course. It requires providers and deployers of AI systems to take measures that support the development of AI literacy of their staff, matched to prior knowledge and the context of use (Regulation (EU) 2024/1689, Article 4). When an AI employee handles a single Vorgang (one intake, one task, one destination), that context fits into five points:

  1. What it does: it classifies documents, extracts metadata and creates a task in exactly one target system. It does not decide on the matter itself.
  2. When it stops: when its certainty falls below the agreed confidence threshold, it hands the case to a named person.
  3. Who takes over: that person in the Fachbereich (the department that owns the Vorgang) assigns the case themselves.
  4. Where it is recorded: every assignment is in the decision log, with timestamp and reasoning.
  5. How to intervene: the human intervention path is documented and can be triggered at any time.

Article 4 asks for measures, not a set level of literacy

Since the Digital Omnibus on AI, Regulation (EU) 2026/1744, the duty reads: providers and deployers take measures to support the development of AI literacy of their staff and of other persons dealing with the operation and use of AI systems on their behalf. They take into account those people's technical knowledge, experience, education and training, the context of use and the persons on whom the system is used. Article 4(1) expressly does not require any specific level of AI literacy for any individual.

The original text of the AI Act asked for measures to ensure, to their best extent, a sufficient level of AI literacy. The duty therefore stays with providers and deployers; its aim is now support rather than a sufficient level. The amendment entered into force on 27 July 2026, the third day after its publication in the Official Journal on 24 July 2026.

Article 3(56) defines AI literacy as the skills, knowledge and understanding that allow an informed deployment of AI systems and awareness of their opportunities, risks and possible harm. A deployer, under Article 3(4), is whoever uses an AI system under its authority, and "operator" is the AI Act's collective term for providers, deployers and others (Article 3(8)). When we say Agentivo operates an AI employee, we mean something narrower: we monitor it, correct it and take responsibility for it. Which role your organisation has under the AI Act is a question for your legal adviser; this page is not legal advice.

The duty has applied since February 2025, and supervision since August 2026

Article 4 sits in Chapter I of the AI Act, and under Article 113(a) Chapter I has applied since 2 February 2025; the Omnibus did not move that date. According to the European Commission, national market surveillance authorities supervise and enforce Article 4 from 2 August 2026.

Article 99(4) does not list Article 4 among the obligations with a fine ceiling set in the Regulation itself. Under Article 99(1), Member States lay down the rules on penalties. Germany's Bundesnetzagentur (in German) notes that a lack of AI literacy can be seen as a breach of the duty of care, particularly where damage results.

What Article 4 does not require: a certificate, a standard course, an AI officer

The Bundesnetzagentur and the European Commission both state what Article 4 does not prescribe:

QuestionAnswerSource
A specific literacy level per person?Not required; Article 4(1) rules it out

AI Act, Article 4

A standardised training course?Not required; formats range from self-study to multi-stage programmes, internal or external

Bundesnetzagentur

A certificate?Not needed; organisations can keep an internal record

European Commission

An AI officer or a governance board?No specific structure is mandated

Bundesnetzagentur, European Commission

Measuring staff knowledge?No obligation to measure

European Commission

What the Bundesnetzagentur does recommend is documentation, and it names three building blocks: assess the need, design the measures, then document, evaluate and refresh them. Conversely, according to the European Commission, simply asking staff to read the instructions for use may in many cases be ineffective. A training course can be one of the measures; it is not the duty itself.

With an operated AI employee, the context is one Vorgang

Article 4 ties the measures to the context in which an AI system is used. For an AI employee that Agentivo builds and operates, that context is narrow: one Vorgang. Which documents belong to it and which classes they are assigned to, we define with your Fachbereich in the process conversation in week 1.

That answers the questions the Bundesnetzagentur uses to assess the need: which people work with which system, for what purpose and with which risks. A typical risk in such a Vorgang is a wrong assignment. The confidence threshold, the agreed hit rate and our corrections in operation address it.

Each role in the Vorgang needs different knowledge

Article 4 asks for each person's knowledge, experience, education and training, and the context of use, to be taken into account. In one Vorgang there are only a few, clearly separated roles:

RoleWhat they should know about the AI employee
Intake, such as the Poststelle (mailroom) or a front officewhich documents belong to the Vorgang and that the AI employee assigns only those
Case workers in the Fachbereichthat the AI employee creates the tasks in the target system, that an assignment can be wrong, and whom to tell when it is
Named people for handed-over casesthat cases below the confidence threshold land with them and that they assign those cases themselves
Management, AI officer, internal auditwhere the decision log is, how to read an entry and who may trigger the intervention path
Data protection officerwhich data the Vorgang touches and what the AVV (data processing agreement) covers
Agentivomonitors, corrects and takes responsibility for operation

Handed-over cases are the core of what people need to learn

Knowledge matters most where the AI employee stops. When its certainty falls below the agreed confidence threshold, it hands the case to a named person. That person should know why the case is with them: the AI employee was not certain enough to assign it, so it stopped.

The person then assigns the case. The matter itself stays with the department: the AI employee does not decide on it, set deadlines or delete anything. According to the European Commission, staff and the people who step in on an AI system need skills targeted to the system they are using.

The decision log makes every assignment explainable

Every assignment is recorded with timestamp and reasoning in the decision log, which serves as the AI audit trail. Whoever answers questions, such as management facing an audit, should know where it is and how to read an entry.

AI literacy here means something concrete: reading an entry and judging whether the reasoning fits the assignment. A complete decision log, available as evidence for EU AI Act audits, is part of every operation.

Everyone knows the intervention path before they need it

Operation includes a defined human intervention path, documented and triggerable at any time. For Article 4 this means the people involved know who may trigger it and when, for example when responsibilities change or wrong assignments pile up.

High-risk AI systems carry an additional requirement. Under Article 26(2) of the AI Act, their deployers assign human oversight to natural persons with the necessary competence, training and authority. For high-risk systems under Annex III, this applies from 2 December 2027 (Article 113(c)). Whether a Vorgang falls into that category is a separate assessment; more on it in the article on EU AI Act compliance for a classification agent.

Your documentation has a fixed reference point in one Vorgang

The Bundesnetzagentur recommends documenting the measures: their type, their scope in content and time, and the people who took part. Regular evaluation and refreshing follow, for example when the system or the context of use changes.

The Pilotbetrieb, our entry engagement, includes a handover document and the decision log; you can draw on both for your records. What else you record is your call.

Wider duties are covered under EU AI Act obligations for companies, data protection under GDPR-compliant AI.

Frequently asked questions

Do we need an AI literacy course with a certificate?

No. Article 4 of the AI Act prescribes neither a format nor a certificate. According to the Bundesnetzagentur, formats range from self-study through workshops to multi-stage programmes, run internally or externally. What matters is that the measures fit the people and the context of use. Record what you did: the type of measure, its scope and who took part.

Does Article 4 apply to small companies and public bodies?

Yes. Article 4 applies to providers and deployers of AI systems, according to the Bundesnetzagentur regardless of sector or size. Under Article 3(4), a deployer can also be a public authority. Article 4(2) adds support for SMEs: the Commission and the Member States support and facilitate their efforts, and the Commission publishes practical examples of how to comply.

Are external service providers covered?

Article 4 covers not only your own staff but also other persons dealing with the operation and use of the AI system on your behalf. The European Commission gives contractors, service providers and clients as examples. With an operated AI employee, Agentivo monitors and corrects operation; who on your side works with it is part of the process conversation in week 1.

What did the Digital Omnibus change in Article 4?

Regulation (EU) 2026/1744 replaced Article 4 and entered into force on 27 July 2026. Previously, Article 4 asked providers and deployers to ensure, to their best extent, a sufficient level of AI literacy. Now they take measures to support the development of AI literacy, without having to reach any specific level for any individual. The duty itself remains.

Are there fines if we do nothing?

Article 99(4) of the AI Act does not list Article 4 among the obligations with a fine ceiling set in the Regulation. Member States lay down the rules on penalties (Article 99(1)), and according to the European Commission, national market surveillance authorities supervise Article 4 from 2 August 2026. The Bundesnetzagentur adds that a lack of AI literacy can count as a breach of the duty of care, especially after damage.

Does Agentivo train our staff?

Training is not part of the Pilotbetrieb. It covers the process conversation with your Fachbereich in week 1, go-live and handover in week 3, a handover document, and then operation: we monitor, correct and take responsibility for the AI employee. How you build on that to support your staff's AI literacy is your call.

The next step

A Pilotbetrieb starts with one Vorgang: one intake, one task, one destination. That also sets the context your Article 4 measures refer to. Fixed price, agreed before we start. Seven weeks, four of them productive operation on your real documents.

Request a Pilotbetrieb

← Back to insights