M.A.I. Consulting
AnalysisEntry8 September 20263 min read

The Omnibus delayed high-risk rules and rewrote Article 4. Neither is a reason to wait

Regulation (EU) 2026/1744 moved the high-risk dates to December 2027 and August 2028 and rewrote Article 4 as a duty to support AI literacy, so AI literacy work is still due but the legal exposure is lighter.

Correction, 14 October 2026. An earlier version of this entry, dated 8 September, described the Digital Omnibus as a proposal and said no proposal moved Article 4. Both statements were out of date: the Omnibus was already law. This version replaces it.

A claim now circulating, usually with visible relief, is that the AI Act has been pushed back and there is now time.

Part of it was pushed back. The part that applies to most mission-driven organisations was rewritten instead.

What the law now says

Regulation (EU) 2026/1744, the Digital Omnibus on AI, was published in the Official Journal on 24 July 2026 and is in force. It does two things that matter here.

It delayed the high-risk obligations. Chapter III duties for Annex III systems now apply from 2 December 2027 instead of 2 August 2026, and for Annex I systems from 2 August 2028 instead of 2 August 2027. If you run a recruitment scoring system, biometric identification or a system that decides access to essential services, that timeline matters to you.

It replaced Article 4. Providers and deployers now have to take measures to support the development of AI literacy among their staff and other people using AI on their behalf. The text adds that this does not require them to guarantee any specific level of AI literacy of any individual. The Commission and Member States are tasked with supporting the effort, and the AI Board with recommendations.

What did not move

Article 4 sits in Chapter I. It has applied since 2 February 2025, and the amendment did not change that date. An organisation using a chatbot to draft donor reports was never in Chapter III. It was in Chapter I, and it still is. Who is in scope is unchanged: deployers, and people acting on their behalf.

What this means for the plan

Two sentences you may have heard are both wrong. "The AI Act was delayed, so we have time" confuses Chapter III with Chapter I. "Article 4 was watered down, so we can drop it" confuses a lighter legal standard with the questions you will actually be asked.

The legal duty is now about effort and proportion: what did you do, and was it sensible for the people and uses involved. A policy, a tool inventory, role-based training and an attendance record are still the plainest way to answer that. The people who ask are a funder writing a due-diligence questionnaire, a board member, an insurer. None of them are waiting for the Official Journal.

Switzerland is outside the EU. A Geneva organisation is reached by the Act through its EU-facing activity, contracts and funding, which is a question to put to your own counsel.

The practical takeaway

Check which chapter the person who said "we have time" was talking about. Then search your own documents for "sufficient level of AI literacy": anything quoting it cites superseded text.

Sources

Series · The EU AI Act, as it now stands · part 3 of 3
02 ยท AI Use Policy

What are our people allowed to do, and how?

If this is the question on your desk, a thirty-minute call tells you whether the service fits, or that you do not need us yet.