A long-awaited text, finally adopted
After months of negotiation, the Digital Omnibus on artificial intelligence has cleared its final stage: approved by the European Parliament on 16 June 2026, it was adopted by the Council on 29 June. This adjustment regulation changes the application timeline of the AI Act, the European regulation on artificial intelligence that entered into force in 2024.
Its stated goal: to give businesses and authorities time to have the technical standards and compliance tools in place before the heaviest obligations apply.
What is postponed
- High-risk systems listed in Annex III (recruitment, education, access to essential services, etc.): their obligations move from 2 August 2026 to 2 December 2027.
- High-risk systems embedded in regulated products (Annex I): from 2 August 2027 to 2 August 2028.
- High-risk systems intended for public authorities get until 2 August 2030.
What does not move
The postponement does not cover everything. The transparency obligations of Article 50, supervision by the European AI Office and the governance rules remain set for 2 August 2026. Only the technical marking of generated content benefits from limited tolerance, for systems already on the market.
In other words: within a few weeks, telling users that they are dealing with an AI, or that a piece of content was generated, becomes an applicable obligation. We will come back to it in detail before the deadline.
The text’s other adjustments
The Digital Omnibus is not limited to the timeline. It also adjusts several AI Act mechanisms, with the stated aim of easing the burden on businesses, particularly smaller ones. National regulatory sandboxes, which allow an AI system to be tested under an authority’s supervision, see their deadline pushed to August 2027. The obligations for general-purpose AI models, applicable since August 2025, remain unchanged.
Who is really affected by “high risk”?
The “high-risk” category is often misunderstood. It does not target generative AI as such, but certain uses: screening job applications, assessing students, granting credit or social benefits, managing critical infrastructure. An internal assistant that helps your teams write or find a document does not fall under it. The same tool used to rank candidates, however, can tip into that category.
Hence the importance of reasoning use by use, rather than tool by tool.
A delay to use, not to endure
The risk with a postponement is complacency. Companies that already use AI, often without having decided to, have every interest in using this time well:
- List existing uses, including undeclared ones (the well-known shadow AI).
- Qualify each use under the AI Act: simple assistant, transparency, or high risk.
- Document the choices: which models, which data, what human oversight.
AI deployed on your own infrastructure makes this easier: you know which model is running, on which data, and you keep a record of how it is used. That is also the spirit of our AI audit, which builds GDPR and AI Act compliance into the mapping of uses from the outset.
Sources: Digital Omnibus regulation on AI, adopted by the Council of the EU on 29 June 2026; summaries published by several law firms (Gibson Dunn, Orrick).
