Updated 25 July 2026: this article originally described a procedural snapshot as of June 2026, when the Omnibus was agreed but not yet law. That has since been resolved, see the update in each section below.
Agreed, and now adopted
On 7 May 2026 the European Parliament and the Council reached a provisional political agreement on the Digital Omnibus on AI, the first amending package for the EU AI Act, and on 13 May the Council published the agreed compromise text. That political agreement has since completed the full legislative process: Parliament approved the text on 16 June 2026, the Council gave final approval on 29 June 2026, and the Omnibus was published in the Official Journal on 24 July 2026, entering into force on 27 July 2026.
The steps that ran
Four things had to happen, in order, and all four are now complete. The European Parliament approved the text in plenary on 16 June 2026. The Council then formally adopted it on 29 June 2026. The text underwent legal-linguistic revision and was signed on 8 July 2026. It was then published in the Official Journal of the European Union on 24 July 2026 and entered into force on 27 July 2026, three days later, ahead of the original 2 August 2026 deadline that the amendments were designed to beat.
Why the sequencing matters
The Omnibus was published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026, amending Regulation (EU) 2024/1689. Legal analysts had been precise on this point while adoption was pending: the new dates would bind only once published. That publication has now landed, so 2 December 2027 is the confirmed high-risk date for stand-alone Annex III systems, not 2 August 2026. Compliance programmes built against the original August 2026 date should re-baseline now.
What changes once it is law
The headline is the staggered deferral: high-risk obligations move to 2 December 2027 for stand-alone Annex III systems and to 2 August 2028 for AI embedded in Annex I regulated products. Transparency obligations under Article 50 still apply from 2 August 2026, with limited additional time, to 2 December 2026, for certain machine-readable content-marking duties. The package also adds a new Article 5 prohibition on AI tools that generate non-consensual intimate imagery or child sexual abuse material, extends documentation simplifications to SMEs and small mid-caps, expands regulatory sandboxes including an EU-level sandbox, and reinforces the AI Office's central role for general-purpose AI oversight. The GPAI obligations in force since August 2025 are not reopened. Our May analysis, what the May 2026 delays mean, explains why the deferral is a re-sequencing rather than a reprieve, and the EU AI Act timeline tracker carries the consolidated dates.
Enforcement is being built while the clock moves
The deferral has not paused the machinery. On 19 May 2026 the Commission opened feedback on draft guidelines for classifying high-risk AI systems, the interpretive layer organisations will be assessed against, and on 1 June 2026 it announced independent expert support for AI Act enforcement. The signal for governance teams is straightforward: the obligations are moving later and getting better tooled, not getting smaller.
What to do in the meantime
Three moves. Update risk registers to the confirmed 2 December 2027 date; the Official Journal publication has landed and August 2026 no longer applies to stand-alone Annex III systems. Use the extra runway on the things the deferral was granted for: inventory and classification against the draft guidelines, data governance, human oversight design and technical documentation. And keep Article 50 transparency work on its original schedule, because it was never deferred.
Sources
European Commission, AI Act policy hub (May to June 2026 updates) · White & Case, EU agrees Digital Omnibus deal (May 2026) · Gibson Dunn, EU AI Act Omnibus agreement (May 2026) · Covington, Inside Privacy: EU AI Act update (May 2026) · Regulation (EU) 2024/1689 (EUR-Lex). General information, not legal advice; verify against the official texts.
