What Article 4 said until July 2026
Regulation (EU) 2024/1689, known as the AI Act, devotes its Article 4 to ‘AI literacy’. In its original version, it required providers and deployers of AI systems to take measures to ensure, to their best extent, a sufficient level of AI literacy of their staff and of other persons using these systems on their behalf.
This article sits in Chapter I of the Regulation, on general provisions. Chapter I has applied since 2 February 2025, well before most of the text. That is the origin of a claim that has become common in the training market: AI training is supposedly mandatory for all employees.
The AI Act defines AI literacy as the skills, knowledge and understanding that allow an informed deployment of AI systems, and awareness of the opportunities and risks they involve.
What the new Article 4 says
Regulation (EU) 2026/1744 of 8 July 2026, presented as the ‘digital omnibus’ package on AI, replaced Article 4. It was published in the Official Journal of the European Union on 24 July 2026 and enters into force on the third day following its publication.
The new text keeps the same scope of persons, but changes the nature of the obligation. Providers and deployers must now take measures to promote the development of AI literacy among their staff, taking into account their technical knowledge, experience, education and training, and the context of use. A new sentence rules out any strict reading: the obligation does not require them to ensure a specific level of AI literacy for any individual.
A second paragraph organises public support. The Commission and the Member States support the efforts of providers and deployers, in particular SMEs. The Commission publishes practical examples of how to comply with the obligation.
| Point | Original Article 4 | Article 4 as rewritten in 2026 |
|---|---|---|
| Verb of the obligation | Ensure, to their best extent | Promote the development |
| Benchmark | A sufficient level of AI literacy | No specific level to be ensured for any individual |
| Scope | Reach a level, to their best extent | Take measures suited to the context, with no guaranteed level |
| Public support | Not provided for in the article | Commission and Member States, priority to SMEs; practical examples published by the Commission |
Why the legislator changed the text
Recital 8 of the 2026 Regulation gives the reason. Stakeholder experience showed that imposing strict obligations on the level to be reached did not suit all types of providers and deployers, and that these obligations created an additional burden, particularly for small businesses.
The same recital adds a sentence every business would do well to read: AI literacy should be a strategic priority, irrespective of regulatory obligations and potential penalties. In other words, the legislator eases the legal constraint without calling the objective into question.
Who is concerned
Article 4 is addressed to two categories: providers of AI systems and their deployers. The AI Act defines a deployer as a natural or legal person, public authority or other body using an AI system under its own authority, except where it is used in the course of a personal non-professional activity.
In our reading, a company that makes an AI assistant available to its teams for their work falls within this definition. An engineering consultancy that has site survey reports summarised, an installer that drafts its quotes with an assistant, an operator that analyses load curves with an AI tool: all are deployers, whatever the risk level of the system used.
What it changes for a business in the energy sector
The obligation remains, its form is free
Nothing imposes a training duration, an official syllabus or a certificate. The obligation concerns measures, suited to the context: who uses which tool, for which tasks, with which data. A short training session on a team's actual uses, a procedure for human review of AI outputs, or a clear rule on the data that must not be entrusted to an online assistant are all possible measures.
Documenting remains the best protection
The text does not require it in so many words. But an obligation of means is assessed on the means deployed: a company that can show what it has done, for whom and when, is in a far stronger position than one that can produce nothing. A simple register of uses and actions is enough to get started.
High-risk systems follow a different timetable
Article 4 is not the only provision that matters for the energy sector. Annex III of the AI Act classifies as high-risk the AI systems used as safety components in the management and operation of the supply of water, gas, heating or electricity. The 2026 Regulation postponed the application of the corresponding obligations to 2 December 2027 for Annex III systems, and to 2 August 2028 for those falling under Annex I. A heating network operator or a facility manager considering AI-driven automated control should build this timetable into the project from the design stage.
Three questions before choosing an action
An obligation of means is not met by a training catalogue chosen at random. In an energy business, three questions are enough to frame the subject.
What data passes through the AI tools? A site survey contains photos of homes, addresses and sometimes occupants' names. A sizing report commits the liability of the person who signs it. Knowing what is entrusted to an online assistant, and what must never be, is the first skill to build.
Who reviews what the AI produces? An assistant can summarise a report, draft a letter or offer a first reading of a regulatory sheet. It can also be confidently wrong: an outdated version of a sheet, a badly converted unit, an invented reference. Every use must have an identified reviewer who knows what they are checking.
Which uses bring a real gain? Repetitive, well-bounded tasks lend themselves best to AI: formatting, extracting information from a document, checking the consistency of a table of measurements. Technical decisions, on the other hand, remain human. In our practice, AI helps spot inconsistencies in a site survey; the sizing itself is still drawn up and validated by a person.
The answers to these three questions outline the training plan: who to train, on which tools, with which rules.
‘AI training is mandatory’: an argument to stop using
Presenting AI training as a legal obligation today means relying on the previous version of the text. For a provider that sells it, it is an inaccurate claim about the reason to buy. For a company that buys it, it is the wrong reason to choose: the right question is not ticking a regulatory box, but knowing what risks its teams take when they use AI without a method.
An accurate wording fits in one sentence: Article 4 of the AI Act, as amended in 2026, requires businesses that use AI systems to take proportionate measures to develop the AI literacy of their teams, without requiring them to ensure a given level.
What we recommend
Start from actual uses. List the AI tools used in the company, by whom and for what, before choosing any training action.
Train on the team's own files. A session built on the real documents of an engineering consultancy or an installer produces reflexes that can be used the next day; a generic course, far less so. That is the principle behind our AI training courses, including Everyday AI to get started.
Write down what has been done. One page per action: date, audience, content, tool concerned. This is the record that shows an obligation of means has been taken seriously.
Follow the Commission's practical examples. The new Article 4 provides for their publication: they will set the level of requirement expected in practice.
To build a plan suited to your teams, write to us. The text cited here is that of Regulation (EU) 2024/1689, as amended in July 2026.
