Key points

  • Article 4 of the AI Act (Regulation (EU) 2024/1689) has applied since 2 February 2025 to all providers and deployers of AI systems, regardless of company size or risk class.
  • Since 27 July 2026 the wording introduced by Regulation (EU) 2026/1744, the Digital Omnibus on AI, applies.
  • The obligation has not been removed. Companies must still take measures that support the development of AI literacy.
  • New is the clarification that companies do not have to guarantee a specific level of AI literacy for any individual.
  • Article 4 does not require a certificate. A simple, traceable record of the measures taken is the sensible approach.

What Article 4 AI Act requires

Article 4 addresses providers and deployers of AI systems. A deployer is any company that uses an AI system under its own authority in a professional context. That already includes companies whose employees use an AI assistant to draft texts, summarise meetings or prepare offers. There is no exemption for small companies or for low-risk systems.

The regulation defines AI literacy as the skills, knowledge and understanding that allow people to use AI systems in an informed way and to be aware of the opportunities, the risks and the possible harm (Article 3(56)). It is not about a technical degree but about confident, informed use in one’s own field of work.

The measures cover a company’s own staff and other persons who deal with AI systems on its behalf, for example freelancers or service providers. They should take into account the technical knowledge, experience, education and training of these people, the context in which the system is used and the persons or groups on whom it is used.

What the Digital Omnibus changed

Regulation (EU) 2026/1744 was published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026. It replaces Article 4. Until then, providers and deployers had to take measures to ensure, to their best extent, a sufficient level of AI literacy of their staff. Now they have to take measures to support the development of AI literacy.

The new wording states explicitly that the obligation does not require providers or deployers to guarantee a specific level of AI literacy for any individual. The yardstick moves from the result for each person to the measures the company takes.

Recital 8 gives the reasoning: strict requirements for a sufficient level of literacy do not suit all types of providers and deployers and burden smaller companies in particular. At the same time it notes that AI literacy should be a strategic priority regardless of legal obligations and possible sanctions.

Two paragraphs are new. The Commission and the Member States are to support companies, and SMEs in particular, in meeting the obligation; the Commission publishes practical examples on a central information platform. The AI Board adopts recommendations with common objectives for promoting AI literacy.

The addressees, the application date of 2 February 2025 and the criteria for tailoring the measures remain unchanged. A company that has done nothing so far is not relieved by the Digital Omnibus.

Who should be trained

Article 4 names neither roles nor a number of hours. Depth depends on the task: drafting e-mails with an AI assistant requires different knowledge than being responsible for AI-supported assessments of applications, loans or customer data. In practice, three levels work well.

  • Everyone who uses AI systems: how they work and where they fail, typical errors such as invented facts, handling confidential and personal data, the internal rules.
  • Subject-matter owners who review and approve results: checking sources, recognising bias, labelling results, responsibility for approval.
  • Managers and project leads: selecting and introducing AI systems, risk classification, the role as provider or deployer, documentation.
  • External people working on the company’s behalf: the same rules as for staff, related to their task.

Which measures make sense

One mandatory course for everybody is rarely the best answer. The wording asks for measures that fit prior knowledge and context of use. This leads to five steps.

  • Take stock: which AI systems are used where and for what, including those built into standard software?
  • Assign roles: who uses, who reviews, who decides?
  • Teach the basics to everyone and go deeper by role, with examples from daily work.
  • Provide a short internal AI policy: permitted tools, handling of data, review and approval steps.
  • Keep it current: include new employees and adapt the content when systems or purposes change.

How to document it

Article 4 prescribes neither a certificate nor a format. The European Commission already explained for the original wording that internal records of training and other measures can be sufficient. With the new wording it matters even more that a company can show which measures it has taken and why they fit its systems and roles.

Clean documentation helps in three situations: questions from a supervisory authority, customer audits and tenders, and when it has to be clarified after an error whether the company met its duty of care. It does not need to be elaborate. It needs to be complete, dated and easy to find.

Template: what belongs in an AI training record

You can adopt the following template as a table or form. One entry per measure is enough; the list of participants is attached.

  • Company, responsible unit and date of the measure
  • AI systems concerned and their purpose
  • Target group and role, for example users, subject-matter owners, managers, external staff
  • Content and learning objectives, matched to prior knowledge and context of use
  • Format and scope, for example online course, workshop, short video with comprehension questions
  • Participants with confirmation of attendance
  • Result of a comprehension check, if one was carried out
  • Version of the training material and of the internal AI policy
  • Date of the next review and triggers for an update

How Yoomind supports this

Most companies already have the content for AI training: the internal AI policy, approval processes, examples from the departments. The Yoomind Workbench turns these approved documents into learning modules, short videos, information pages and comprehension questions, multilingual and in the company’s design. Experts review every result before it is used.

With the Yoomind All-in-One learning platform, modules can be assigned to target groups; learning status, history and completions are documented. Training and record come from the same source, and when the policy changes both are updated together.

Sources and status

As of 10 November 2026. This article is an overview and does not replace legal advice. The wording of the regulations in the Official Journal of the European Union is authoritative.

  • Regulation (EU) 2024/1689 (AI Act), Article 3(4) and (56), Article 4, Article 99, Article 113
  • Regulation (EU) 2026/1744 of 8 July 2026 (Digital Omnibus on AI), OJ L, 2026/1744, 24.7.2026: new Article 4 and recital 8
  • European Commission, AI literacy questions and answers (2025)

Frequently asked questions

Did the Digital Omnibus abolish the AI training obligation?+

No. Article 4 of the AI Act still applies. Since 27 July 2026 providers and deployers must take measures that support the development of AI literacy of their staff. Only the benchmark of ensuring a sufficient level of literacy has been dropped.

Does Article 4 AI Act apply to small companies?+

Yes. Article 4 has no exemption based on company size. The measures may match prior knowledge and context of use, so they can be lean where a company uses few, simple AI applications. The Commission and the Member States are explicitly asked to support SMEs.

Is a one-off training enough?+

The regulation does not set an interval. Because AI systems, purposes and teams change, a single training is usually not enough. A basic course at the start, deeper modules by role and a review whenever new systems are introduced make sense.

Do employees need a certificate?+

No. Article 4 requires neither a certificate nor a particular training format. It is enough to document the measures in a traceable way: content, target group, date, participants and version of the material.

Are there fines for breaching Article 4?+

Article 99 of the AI Act lists no specific fine for Article 4. Member States do, however, lay down penalties and other enforcement measures, and missing training can matter for liability after damage has occurred. Have the individual case checked by a lawyer.

What does an AI training record look like?+

An AI training record states for each measure which AI systems are concerned, who was trained, what was taught, when it happened and which version of the material was used. A table with the list of participants attached is sufficient.