EU AI Act Article 4: original and new wording

By Marco Kohns, Co-founder of ENLIX, lecturer in AI and growth

· 8 min read

Article 4 of the EU AI Act governs AI literacy. Since 27 July 2026 its first paragraph reads: providers and deployers of AI systems take measures to support the development of AI literacy of their staff and other persons acting on their behalf, without having to guarantee a specific level for anyone. The version before applied from 2 February 2025 to 26 July 2026 and required ensuring, to their best extent, a sufficient level of AI literacy, which is the old wording. Here are both versions verbatim, what changed and how it came about.

What it means in practice, who is affected, whether sanctions apply and how to document training, is covered in our article on the AI literacy obligation, and the term itself, with its definition and the levels, in our article on AI literacy. This one is the reference.

What does Article 4 of the EU AI Act say today?

Since 27 July 2026, Article 4 has three paragraphs. The first addresses organisations that offer or use AI, the second the Commission and the Member States, and the third the EU's AI Board. The wording, as set by Article 1(5) of Regulation (EU) 2026/1744:

Article 4 AI literacy

  1. Providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf, taking into account their technical knowledge, experience, education and training and the context the AI systems are to be used in, and considering the persons or groups of persons on whom the AI systems are to be used. This obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual.

  2. The Commission and the Member States shall support and facilitate the efforts of providers and deployers of AI systems, in particular SMEs, in fulfilling their obligation under paragraph 1 of this Article. For that purpose, the Commission shall publish practical examples of how to comply with that obligation on the single information platform referred to in Article 62(3), point (b).

  3. The Board shall adopt recommendations, taking into account European competence frameworks, to support the Commission and Member States in the promotion of AI literacy required under paragraph 1, including by setting out common objectives.

The definition of AI literacy in Article 3(56) is unchanged: the skills, knowledge and understanding to make an informed deployment of AI systems and to be aware of their opportunities, risks and possible harm.

What did the original Article 4 say?

The original version was a single paragraph and applied from 2 February 2025 to 26 July 2026:

Providers and deployers of AI systems shall take measures to ensure, to their best extent, a sufficient level of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf, taking into account their technical knowledge, experience, education and training and the context the AI systems are to be used in, and considering the persons or groups of persons on whom the AI systems are to be used.

Many German pages still present this text as current law. In our sample of 13 October 2026 it was 22 of 46 pages ranking on google.de for the main German search terms, including pages with an update date after 27 July 2026. By type of publisher:

German pages presenting the original wording as current law, by type of publisher

Training (18)9 pagesLaw firms (7)5 pagesChambers (4)3 pagesConsultancies (6)2 pagesBanks (2)2 pagesMedia (1)1 pages

Of 41 pages that cover Article 4, from the first 20 results on google.de for four German search terms, checked on 13 October 2026. Number of pages checked per group in brackets; Training means AI training providers, Chambers the German chambers of commerce. Our own analysis, method in the article on the AI literacy obligation.

German pages presenting the original wording as current law, by type of publisher
Training (18)9 pages
Law firms (7)5 pages
Chambers (4)3 pages
Consultancies (6)2 pages
Banks (2)2 pages
Media (1)1 pages

What exactly changed in Article 4?

The goal of the obligation changed, not who it addresses. Both versions address providers and deployers and the same people. In detail:

ElementOriginal wordingNew wording
Who is obligedProviders and deployersProviders and deployers, unchanged
For whomStaff and other persons on their behalfunchanged
What to doTake measures to ensure a levelTake measures to support development
Target level"sufficient level", "to their best extent"none, expressly no guaranteed level
What to take into accountKnowledge, experience, education, context, affected personsunchanged
Role of the statenot coveredCommission and Member States support, especially SMEs (para. 2)
Examplesnot coveredCommission publishes practical examples (para. 2)
AI Boardnot coveredadopts recommendations (para. 3)
Article 4 AI Act, the two versions compared element by element.Source: OJ L 2024/1689, OJ L 2026/1744

The reasoning is in recital 8 of the amending regulation: stringent obligations to ensure a sufficient level of AI literacy would not suit all types of providers and deployers and create extra burden especially for smaller companies, which is why the old approach was replaced. AI literacy should be a strategic priority regardless of legal obligations.

How did the amendment come about?

The change is part of the Digital Omnibus on AI, a package that simplified several rules of the AI Act and postponed the high-risk obligations. For Article 4, two different models were on the table at the start and at the end:

Article 4 in the legislative procedure

obligation sits with the organisation

Commission proposal, 19 November 2025

  1. Commission and Member States encourage providers and deployers to take measures
  2. the target of a sufficient level stays in the text
  3. the obligation sits with the state

0 of 3 points with the obligation on the organisation

Adopted wording, in force 27 July 2026

  1. providers and deployers take measures
  2. they support the development of literacy
  3. no one has to assure a specific level
  4. Commission and Member States support them

3 of 4 points with the obligation on the organisation

Proposal COM(2025) 836, Article 1(4); adopted wording OJ L 2026/1744, Article 1(5).

The later steps, all from the Official Journal of the amending regulation: position of the European Parliament of 16 June 2026, Council decision of 29 June 2026, signature in Strasbourg on 8 July 2026, publication on 24 July 2026. The regulation entered into force on the third day after publication, 27 July 2026, and sets no later date for the new Article 4.

The Commission's AI literacy Q&A says the amendment entered into force in mid-July. The Official Journal is authoritative here, not the Q&A.

What do paragraphs 2 and 3 mean?

Paragraphs 2 and 3 do not address organisations. Paragraph 2 obliges the Commission and the Member States to support providers and deployers, SMEs in particular, with paragraph 1. The Commission is to publish practical examples on the single information platform under Article 62(3)(b), which it runs as the AI Act Service Desk.

A collection of such examples has existed for a while: the Commission's living repository of practices from organisations that signed up to the AI Pact. The Commission states that replicating those practices does not automatically grant presumption of compliance with Article 4. Paragraph 3 tasks the AI Board, the body of Member States at EU level, with adopting recommendations, taking European competence frameworks into account. An overview of these and other free resources is in AI literacy: free resources.

What applies on top for high-risk AI?

For high-risk AI there is a separate rule the Omnibus left unchanged. Article 26(2) of the AI Act requires deployers of such systems to "assign human oversight to natural persons who have the necessary competence, training and authority, as well as the necessary support". Here a concrete competence requirement for specific people remains.

After the postponement by the Omnibus, the obligations of deployers of high-risk AI apply from 2 December 2027 for Annex III systems and from 2 August 2028 for Annex I systems. For most offices that use language models for text, research or code, Article 4 is the rule that counts today. The AI training record generator and the AI literacy test show how to put it into practice.

Frequently asked

What does Article 4 of the EU AI Act say?

Since 27 July 2026: providers and deployers of AI systems take measures to support the development of AI literacy of their staff and other persons acting on their behalf. No specific level of literacy has to be guaranteed for anyone. Paragraph 2 obliges the Commission and the Member States to support organisations, and paragraph 3 the AI Board to adopt recommendations.

When was Article 4 amended?

By Regulation (EU) 2026/1744, signed in Strasbourg on 8 July 2026, published in the Official Journal on 24 July 2026 and in force since 27 July 2026.

Does Article 4 apply only to high-risk AI?

No. Article 4 applies to providers and deployers of all AI systems, whatever the risk class. For deployers of high-risk AI, Article 26(2) adds that human oversight is assigned to people with the necessary competence. Those duties apply to Annex III systems from 2 December 2027.

What had the Commission proposed for Article 4?

In the proposal of 19 November 2025, the Commission and the Member States would only have encouraged organisations to take measures, which would have moved the obligation to the state. The legislator did not adopt that and kept the obligation with organisations.

Written by

Marco Kohns

Co-founder of ENLIX, lecturer in AI and growth

Marco worked as a growth product manager at a Silicon Valley scale-up and has been teaching that way of working ever since. Today he runs ENLIX with Tobias and builds two products of his own on the same systems, which is what the courses open up.

  • Growth product manager at a Silicon Valley scale-up, Series A to B, backed by a16z, General Catalyst and Sapphire, with users in over 100 countries and more than 20,000 cities
  • Peer-reviewed research in the Journal of Business Research on generative AI in growth, with Prof. René Bohnsack. The research began in summer 2022, months before ChatGPT was public
  • Executive education lecturer at Católica-Lisbon, over 10 seminars, more than 1,500 people taught

Keep reading