AI literacy after the Omnibus: a softer duty that still needs a record

AI literacy after the Omnibus: a softer duty that still needs a record

One sentence in the EU AI Act changed on 27 July 2026, and the change is larger in effect than it looks on the page.

Article 4, the AI literacy article, has applied since February 2025. In its original form it said that 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 of others operating AI systems on their behalf. The amending regulation known as the Digital Omnibus on AI, Regulation (EU) 2026/1744, replaced that wording. Article 4 now says providers and deployers “shall take measures to support the development of AI literacy”. It then adds a sentence that was not there before: “This obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual.”

I work in technology training, so a strong reading of Article 4 would suit me. The amended text does not support one, and I would rather say so plainly than let anyone buy training on the strength of a duty that has been softened.

What changed, and what did not

Three things are new. The duty is to support the development of AI literacy, where it used to be to ensure a sufficient level of it. There is an express statement that no particular level has to be guaranteed for any person. And the article now gives the Commission and the Member States a role: they are to support providers and deployers, small and medium-sized enterprises in particular, and the Commission is to publish practical examples of how to comply on the AI Act’s single information platform. A new third paragraph asks the European Artificial Intelligence Board to adopt recommendations that take account of European competence frameworks.

Much stayed the same. The duty still falls on providers and deployers. It still covers their staff and other people dealing with the operation and use of AI systems on their behalf, which brings in contractors and suppliers acting for an organisation. And the measures are still meant to reflect people’s technical knowledge, experience, education and training, the context the systems are used in, and the people or groups affected by them.

The amending regulation entered into force on the third day after its publication in the Official Journal on 24 July, so 27 July 2026, and I found no later application date for the new Article 4. It is a regulation, so it applies directly in every Member State, including Germany, Austria, the Netherlands, Belgium and Luxembourg, without national transposition.

Who it reaches outside the EU

UK and Swiss organisations are not exempt by virtue of being outside the EU. Article 2 applies the Act to providers placing AI systems on the EU market or putting them into service there, wherever those providers are established, and to providers and deployers in third countries where the output produced by the AI system is used in the Union. A UK company that sells an AI-enabled product into the EU, or a Swiss firm whose AI-generated output is used by EU customers, needs to read Article 2 with its own facts in front of it. This is a summary of the scope article, not legal advice, and the edge cases deserve a lawyer’s time.

What “support” asks for in practice

A duty to take measures, without a guaranteed outcome, still has to be met with something. The honest question for any organisation is what it did, for whom, and whether that fitted the systems those people use.

That points towards a record of measures, proportionate to role and context, more than towards a single course for everyone. An all-staff module produces a completion rate. It says little about whether the people configuring AI into a process understand it any better than the people using a chat assistant to draft an email.

A proportionate approach would look different for different groups. My suggestions, which are not requirements of the Act, would be these.

  • Everyone using a general-purpose AI assistant needs a short briefing on which tools are approved, what data may go into them and how to check what comes out.
  • Teams building AI into a process need training on the system’s limits, how it was tested and how it fails.
  • People who approve or oversee decisions supported by AI need enough depth to challenge the output and to know when to stop using it.
  • Suppliers and contractors operating AI on the organisation’s behalf need the same rules in their contracts, and evidence that they were briefed.

For each group, the record would show the system or tool, who uses or operates it, their starting point, the measure taken, the date, the evidence, and what would trigger a review, such as a new model version, a new tool or a complaint.

The case for doing less, and why I would not

The obvious response to a softer duty is to do less, and for some organisations that is reasonable. An organisation that spent heavily on generic training in 2025 can now right-size it, and should. Nothing in the amended text requires every member of staff to hold an AI certificate, and I would be wary of anyone who says it does.

The case against doing nothing is practical as much as legal. The article still says “shall take measures”, and national market surveillance authorities supervise the Act. More to the point, the people most likely to cause harm with an AI system are the ones who use it confidently without understanding its limits. That risk did not change on 27 July. The paperwork just became lighter.

There is also a timing point. The Commission’s practical examples, and the Board’s recommendations, will shape what supervisors expect. An organisation with a short, honest record of what it did will find it easier to adjust to that guidance than one with nothing on file.

What I would do this month

List the AI systems and tools in use, including the ones staff adopted without asking. For each, name the people who use or operate it and decide what they need to know. Take the measures that fit, write down what you did, and set a trigger for reviewing it. Keep it short enough that someone will actually update it.

For most organisations that is a smaller task than an all-staff programme, and a more useful one.


Sources, checked on 4 November 2026:

  • Regulation (EU) 2024/1689 (the AI Act), Official Journal L, 12 July 2024: original Article 4 (“shall take measures to ensure, to their best extent, a sufficient level of AI literacy”) and Article 2(1)(a) to (c) on scope. https://eur-lex.europa.eu/eli/reg/2024/1689/oj
  • Regulation (EU) 2026/1744 of 8 July 2026 (Digital Omnibus on AI), Official Journal L, 24 July 2026: Article 1(5) replacing Article 4; Article 4 on entry into force “on the third day following that of its publication” and direct applicability in all Member States. https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=OJ:L_202601744
  • This essay is commentary, not legal advice.