The vote
The European Union’s flagship artificial-intelligence law received its first major adjustment since it was passed in 2024. On 16 June 2026, the European Parliament backed the Digital Omnibus on AI in plenary, on the basis of the provisional agreement reached on 7 May between Parliament and the Council. The Internal Market (IMCO) and Civil Liberties (LIBE) committees had already endorsed it on 2 June by 93 votes in favor, 4 against and 15 abstentions.
It is worth marking the exact status, because it is not yet final. What remains is formal adoption by the Council of the Union, expected before 2 August 2026; only after publication in the Official Journal will the new dates be firm, and the regulation will enter into force three days later.
The hand that loosens
The bulk of the package is a postponement. Obligations for the use-based high-risk systems of Annex III — covering, among others, recruitment, credit-scoring, education, border-control and law-enforcement tools — are pushed from 2 August 2026 to 2 December 2027, a sixteen-month delay. Those for AI embedded in regulated products under Annex I — medical devices, machinery, vehicles — move from 2 August 2027 to 2 August 2028. The requirement that each member state establish at least one national regulatory sandbox also moves to 2 August 2027.
Transparency got a shorter reprieve. For generative systems already on the market before 2 August 2026, the obligation to mark synthetic content in a machine-readable way is deferred to 2 December 2026; products placed on the market after that date must comply from day one.
The hand that tightens
On one point, the Omnibus does not loosen: it tightens. The agreement adds to Article 5 a prohibition on AI systems that generate or manipulate intimate images, video or audio of an identifiable person without their explicit consent — the so-called “nudify” apps — and on those that produce child sexual abuse material (CSAM), effective from 2 December 2026. As several firms that followed the negotiation noted, this prohibition was not in the Commission’s text: Parliament pushed it through, and it sits in the regulation’s top penalty band, up to 35 million euros or 7 percent of global turnover. For providers, the ban reaches not only systems intended for such use, but those in which such generation is a reasonably foreseeable and reproducible outcome without significant modification and which lack adequate technical safeguards.
The shortcut that closes
There was an attempt to lighten the load that did not prosper, and it is worth noting. The Commission had proposed removing the obligation to register, in the EU database, systems operating in Annex III contexts that their providers self-assess as not high-risk; Parliament and the Council rejected the deletion and kept the registration in a streamlined form. In practice, a self-assessment that used to be an internal memo becomes a public filing: whoever decides that their HR or credit tool is not high-risk will have to defend that classification before a searchable list available to authorities.
Two readings
The same package admits two narratives, and both are on the table. The Council presented the agreement as support for business that cuts recurring administrative costs and brings legal certainty and more harmonized application, in line with the Union’s simplification and competitiveness agenda. On the other side, several analysts warned that the most debatable point is the deferral of the sandboxes, precisely the tool the small players the rule says it wants to protect need most, so delaying it weakens a pro-innovation instrument. The underlying debate is the one the Draghi report on competitiveness opened: whether Europe can comply with its own regulation without giving up competing.
What remains
The picture, as this report closes, is of a text approved by one chamber and pending in the other. The prohibitions in force since February 2025 — social scoring, subliminal manipulation — still stand, and the rules for general-purpose models are untouched. No fine has been imposed under the regulation as of mid-2026. The reshaped calendar will be final only when the Council adopts it and it is published; until then, the new dates are the planning baseline used by authorities and law firms, not yet firm law.