What the Digital Omnibus Actually Proposes
The EU’s Digital Omnibus is a package of proposed legislative amendments to EU digital regulation, including the AI Act. Trilogue negotiations, the three-way process between the European Commission, Parliament, and Council, are underway as of April 2026, confirmed by A&O Shearman’s direct coverage of the negotiations. The proposals under discussion would affect three separate compliance tracks.
Track 1, Standalone high-risk AI systems: Legal analysts report the Digital Omnibus proposes a new compliance deadline of December 2, 2027. The original deadline is August 2, 2026. That’s a 16-month extension.
Track 2, High-risk AI embedded in regulated products: For AI systems integrated into products already subject to EU product safety law, medical devices, machinery, vehicles, the proposed deadline extends further, to August 2, 2028. That’s a 24-month extension from the original August 2, 2026 date.
Track 3, Watermarking: The Digital Omnibus reportedly proposes extending the watermarking compliance deadline to November 2, 2026, a more modest extension from the original timeline.
Every one of these figures comes from law firm analysis, specifically reporting by Jones Walker and Fladgate, not from official EU Commission documents. Until the Commission publishes formal amended deadline language, these numbers should be treated as reported proposals, not enacted law. They can change in negotiation. They can be withdrawn. The November 2, 2026 watermarking figure, in particular, is close enough to current that teams in that compliance track can’t afford to treat it as relief rather than a shifting target.
The Digital Omnibus may also clarify the scope of high-risk AI classifications – potentially narrowing which applications face the full compliance burden. Legal analysts suggest a carve-out for AI systems that perform convenience or optimization functions, rather than consequential decision-making, is under discussion. The precise language remains in negotiation. For organizations whose systems sit near the classification boundary, this scope question is as material as the deadline dates, a narrowed classification could remove the compliance obligation entirely, while a retained classification with a delayed deadline changes the timeline but not the obligation.
What Changes If the Proposals Are Adopted
The practical compliance calendar comparison is the most useful frame for decision-making.
| System Type | Original Deadline | Proposed Deadline | Extension | Status |
|---|---|---|---|---|
| Standalone high-risk AI | August 2, 2026 | December 2, 2027 | +16 months | PROPOSED, PENDING TRILOGUE |
| High-risk AI in regulated products | August 2, 2026 | August 2, 2028 | +24 months | PROPOSED, PENDING TRILOGUE |
| Watermarking | Original timeline | November 2, 2026 | Modest extension | PROPOSED, PENDING TRILOGUE |
If Track 1 proposals are adopted, organizations deploying standalone high-risk AI systems gain 16 months. That’s significant: time to complete conformity assessments more thoroughly, to build technical documentation to a higher standard, to integrate AI governance into existing risk management systems rather than running it as a separate track. Organizations that have been resourcing compliance work on a compressed timeline could reallocate some of that capacity.
If Track 2 proposals are adopted, organizations with AI embedded in regulated products gain the most runway, two full years from the original deadline. Medical device companies, automotive manufacturers, and industrial equipment makers integrating AI into product lines already subject to CE marking requirements have the most complex conformity assessment pathways. Two additional years is meaningful for those organizations.
But “if adopted” is doing significant work in both sentences above.
What Doesn’t Change Regardless of the Delay
The Digital Omnibus proposals, even if fully adopted, don’t touch three existing compliance tracks.
First: The prohibitions on unacceptable-risk AI practices are already in effect and are not subject to the deadline delay proposals. Systems that manipulate users through subliminal techniques, exploit vulnerabilities of specific groups, or enable real-time biometric surveillance in public spaces are already prohibited. No trilogue outcome changes that.
Second: GPAI (general-purpose AI) model obligations are in force under the EU AI Act’s existing timeline and are not within the scope of the Digital Omnibus delay proposals as reported. Organizations deploying or building on GPAI models need to maintain their current compliance tracks on the original schedule.
Third: Internal governance work, documentation frameworks, AI system inventories, risk classification procedures, human oversight protocols, is unaffected by deadline shifts because it should exist independent of external deadlines. Organizations that have built these structures have reduced their compliance risk regardless of which deadline ultimately governs.
The compliance work least affected by a potential delay is also the compliance work with the longest internal lead time. That’s not a coincidence.
The “Continue Preparing” Argument, and Why It Holds
The consensus position among legal analysts tracking the Digital Omnibus negotiations is straightforward: continue planning against the original August 2, 2026 deadline while monitoring the trilogue for formal adoption of the proposed extensions.
The argument rests on asymmetric risk. Consider two scenarios.
Scenario A: You continue preparing for August 2. The proposals are adopted, and the deadline moves to December 2027. You’ve invested compliance resources against a deadline that shifted. The cost is some work completed earlier than strictly necessary. You’re ahead of the new deadline with time to spare. Compliance risk: low.
Scenario B: You pause compliance investment in anticipation of the delay. The proposals are modified in trilogue and the delay is narrowed, or the proposals don’t survive negotiation at all, and August 2 stands. You’ve lost months of preparation time with no recovery path. Compliance risk: high.
Scenario A’s downside is workload efficiency. Scenario B’s downside is regulatory exposure. For organizations that have already invested significantly in August 2 preparation, the marginal cost of continuing is lower than the risk of stopping. For organizations that haven’t started, the trilogue uncertainty is not a reason to delay – it’s a reason to accelerate, because the original deadline is still the legally operative date.
Key Dates to Track
Legal analysts tracking the trilogue projected a political agreement could be reached as early as April 28, 2026. That projection was made before April 11 and should be treated as an estimate, not a scheduled announcement. A political agreement is not the same as formal adoption: it precedes the legal-linguistic review, formal Council vote, and publication in the EU Official Journal. Even an April 28 agreement would not mean the deadline extensions are immediately operative.
The dates that matter in sequence: political agreement announcement (watch for official EU press releases, not secondary reporting); formal Commission publication of amended Regulation text; publication in the Official Journal with effective date; and any transitional provisions that affect organizations already in compliance programs.
If you’re managing EU AI Act compliance for an organization, assign a monitoring responsibility to the EU Commission’s official channels, not law firm blogs. Law firm analysis is useful for interpretation, it’s not a substitute for official publication.
What to Watch
Three specific markers determine whether and when compliance calendars need to be updated.
One: Whether the April 28 political agreement projection holds, and what the announcement says about the specific deadline language. A political agreement that doesn’t specify final deadline dates is a progress signal, not a compliance calendar update.
Two: Whether the high-risk classification scope clarification survives negotiation and what its precise language covers. This matters more for organizations near the classification boundary than the deadline shift does.
Three: Whether the EU Official Journal publishes amended deadline language, this is the only event that actually changes your compliance obligation. Everything before it is a signal to monitor, not a reason to adjust.
TJS synthesis: The Digital Omnibus proposals, if adopted as reported, would represent the most significant adjustment to the EU AI Act’s implementation timeline since the Act’s passage. The relief is real, 16 to 24 months is material. But it’s proposed relief, not enacted relief. The organizations best positioned to benefit from an extended deadline are the ones that continued preparing for August 2 while the negotiations concluded. The compliance posture that survives either outcome, adoption or rejection of the proposals, is the same: build against the original deadline, monitor the trilogue, and update your calendar when the EU Official Journal tells you to, not when a law firm blog does.