The EU AI Act timeline after the Digital Omnibus: what actually changed

If you have been trying to pin down when the EU AI Act’s high-risk obligations actually bite, you have run into a moving target. The headline date most people memorized, 2 August 2026, is no longer the operative one for high-risk systems. Here is the current, sourced picture as of June 2026, and the part that matters: what did not change.

Last reviewed: 21 June 2026. This is an evolving legislative process; verify against primary sources before relying on any date.

What changed

In early May 2026, the Council and the European Parliament reached a provisional political agreement on the “Digital Omnibus on AI,” a package that amends the AI Act to simplify and stagger its rollout. The agreement was announced on 7 May 2026 and confirmed by member-state representatives in the Council on 13 May 2026.

The core effect is a delay to the high-risk obligations:

High-risk systems under Annex III (standalone use cases such as employment, credit, and essential services): applicability moves from 2 August 2026 to 2 December 2027. That is roughly a 16-month deferral.
High-risk AI embedded in regulated products under Annex I: applicability moves from 2 August 2027 to 2 August 2028.

The stated rationale is that compliance depends on technical standards that are not yet finished, and organizations need more runway to meet the obligations.

What did not change

This is the part the “the AI Act is delayed” headlines get wrong. The Omnibus defers the high-risk obligations. Several parts of the Act are already in force and are not affected:

Prohibited practices (the banned use cases) have applied since February 2025.
General-purpose AI model obligations have applied since August 2025.
The Act’s overall structure, governance bodies, and penalty regime remain in place.

So the accurate statement is narrow: the high-risk compliance clock moved, not the whole law.

The caveat that matters

None of this is final yet. A provisional political agreement is not enacted law. The changes take legal effect only once the Omnibus is formally adopted and published in the Official Journal of the EU, which is expected before 2 August 2026. Until then, the current text technically still carries the August 2026 date, which is exactly why you are seeing two different timelines quoted. Treat the new dates as the strong expectation, not a settled fact, and watch for the published text.

What it means for governance planning

More time is not a reprieve. It is preparation time, and the evidence from every compliance cycle is that organizations spend it the wrong way: the deadline moves out, urgency drops, and the work that takes the longest, building a defensible record of what your AI systems actually do, gets deferred again. Then the new deadline arrives with the same gap that existed before.

Three practical points:

Procurement does not wait for the Official Journal. Large buyers are already requiring AI documentation in contracts. The regulatory date is not the only forcing function; your customers are another.
The slow work is the documentation, not the deadline. Mapping what a system does, against what you claim it does, takes weeks to months. A later deadline does not shorten that; it just changes when you should start, which is now.
A delay is the most dangerous moment for the compliance illusion. The risk is not that the AI breaks. It is that the gap between claim and reality widens quietly while the calendar pressure is off, and surfaces when a regulator, acquirer, or customer finally asks.

The right use of the extra time is to get to a defensible position early and hold it, rather than to push the preparation to the edge of a date that has already moved once.

Where to start

If you want a fast read of where your AI systems stand against this, the 10-question diagnostic returns a written tier classification in about ten minutes. If you need a defensible, documented assessment ahead of the high-risk deadlines, tell us about the system and we will scope it.