The Digital Omnibus changes the EU AI Act. But not in the way you think.

You are reading about it everywhere: the EU AI Act is being "simplified". The Digital Omnibus was adopted by the European Parliament and the Council, published in the Official Journal on 24 July 2026 and in force since 27 July 2026, which changes the law on several points. The predictable reaction in many a management meeting: good, that AI file can wait a little longer. That is precisely the wrong conclusion. Let me explain why.

What actually changes

The Digital Omnibus on AI was approved by the European Parliament on 16 June 2026 and received the Council's final green light on 29 June 2026. It was published in the Official Journal on 24 July 2026 as Regulation (EU) 2026/1744 and entered into force on 27 July 2026. The Omnibus covers more than most summaries mention, from a new prohibition on AI that generates non-consensual intimate imagery to adjustments to the high-risk definition. But two changes matter to almost every organisation.

One: the deadlines for high-risk AI shift, depending on the category, to December 2027 or August 2028. Anyone using or providing high-risk systems gets more time.

Two, and this affects everyone: Article 4 on AI literacy has been rewritten. The original text asked organisations to ensure, to the best of their ability, "a sufficient level of AI literacy" among their staff. An outcome-oriented wording: the emphasis was on the level your people were supposed to reach. The new text asks organisations to "take measures to support the development of AI literacy", fitting role, experience and context, and explicitly without any guarantee that a given individual reaches a specific level of knowledge. A deliberate effort obligation, in other words.

What does not change

Article 4 has not been given a new application date: the obligation has applied since 2 February 2025 and continues to apply. Only its wording has changed with the Omnibus. The duty to actually do something about AI literacy does not disappear, it is phrased differently. The prohibited AI practices also remain in place, and one is even added. And while the high-risk deadlines move back, supervision moves the other way: member states are designating their supervisory authorities, and since 2 August 2026 supervision has been operational. We set out separately what that supervision did and did not change on 2 August 2026.

Anyone reading the Omnibus as a stay of execution is reading it wrong. The obligation that touches your entire organisation has not been postponed.

Why an effort obligation is anything but optional

Let's be honest: on the outcome side, this is a relaxation. The guarantee of a knowledge level per employee is gone. But anyone concluding that the duty has become optional is missing where the centre of gravity shifts.

The old standard was strictly worded but hard to pin down in practice. Because what is "a sufficient level"? That had not yet been sharply defined in binding guidance, and that vagueness left room to wait and see. The new text is worded more mildly but is more concrete in what it asks of you: measures. And measures you cannot demonstrate carry hardly any weight as evidence of compliance in supervision or an audit. In practice, your evidential position therefore shifts to documentation: whoever takes measures must be able to show which ones, for whom, when, and why they fit the role and context. For that record-keeping, a two-page AI policy is enough to start with. The key word is no longer "sufficient". The key word is "demonstrable". Our free EU AI Act Compliance Checklist takes you through what "demonstrable" means in practice. If you want to house that documentation in a management system, that goes further than Article 4 alone.

What such measures look like is deliberately not prescribed by the law. No mandatory exam, no mandatory certificate, no fixed format. But in practice, they are no mystery. Think of an inventory of which AI is used in your organisation. Training that fits role and risk, from the front desk to senior management. An exam with a certificate, not because it is required but because it is strong evidence that the measure is serious, verifiable and role-specific. And refreshers whenever the tools or the rules change. Anyone who has that in order can answer any question from a supervisory authority, auditor or insurer with a file instead of a promise.

So the bar does not disappear. It shifts: from what your people have in their heads to what you can show you have done for them.

What this means for your approach

Three things, briefly.

Pause nothing. The obligation has applied since February 2025, and since 2 August 2026 supervision has been running. Whoever starts now is ready before the first questions arrive. Our AI literacy training under the AI Act is built for that, with an exam and a certificate for each member of staff. The price per person is stated openly. We calculated separately what training delivers.

Documentation becomes the core. Every measure you take must be traceable: who was trained, on what, when, with what result. Evidence is the product of your effort.

Keeping it up is part of it. An effort is not a one-off action. New tools, new employees and new rules call for a recurring rhythm. How to set up that rhythm in practice is something I described earlier this week in a separate Insight on AI policy in practice.

The Digital Omnibus is no reason to sit back. It is the legislator writing down more precisely what was always expected of you: show that you take your people seriously in a working environment where AI is not going away. Not sure where you stand? The free AI Adoption Scan maps it out. Would you rather hand it over? We do that in an AI Adoption Assessment. What changes around the AI Act after that we track every working day in the EU AI Act Monitor. And otherwise, you can simply get in touch.

Frequently asked questions

Has Article 4 of the EU AI Act been abolished?

No. Article 4 has applied since 2 February 2025 and continues to apply. The Digital Omnibus has rephrased the duty, from an outcome-oriented wording to a deliberate effort obligation: taking measures to develop AI literacy, demonstrably and fitting role and context.

Does the EU AI Act still apply?

Yes. The Omnibus amends parts of it, shifts the high-risk deadlines to December 2027 or August 2028 and even adds a new prohibition. The law itself remains in force, including the prohibited AI practices and the literacy obligation.

When does the Digital Omnibus take effect?

The Omnibus was adopted in June 2026 by the European Parliament (16 June) and the Council (29 June), published in the Official Journal of the EU on 24 July 2026 and entered into force on 27 July 2026. The amended text of the AI Act has applied since that date.

Rob Ummels, AIAdopt. AI adoption, AI literacy and practical guidance for organisations that want to use AI wisely.

Written by Rob Ummels in collaboration with Claude (Anthropic). Editorial responsibility: AIAdopt.