The Digital Omnibus on AI is a legislative package from the European Commission that was introduced into the ordinary legislative procedure in March 2026. Its goal is to amend several digital legal acts, most notably by postponing the key effective dates of the EU AI Act.
Under the current proposal, high-risk obligations under Annex III would be pushed back from August 2, 2026, to December 2, 2027, and obligations for high-risk systems under Annex I (safety-related products) to August 2, 2028. This postponement is contingent upon the availability of harmonized standards.
Legislative status as of April 20, 2026: The Council adopted its position on March 13, 2026, and the European Parliament on March 26, 2026. The IMCO and LIBE committees voted 101 to 9 in favor of the postponement in their joint report, with 8 abstentions. Trilog negotiations are underway; an agreement before June 2026 is necessary for the postponement to take effect before August 2, 2026.
While a postponement seems politically likely, it has not yet been finalized. For companies with ongoing SAP AI projects, the question is how to plan amidst this uncertainty.
First, the deadline for the Commission’s classification guidance under Article 6 passed on February 2, 2026, without the guidance being issued. Consequently, classification uncertainty remains regardless of the Digital Omnibus. Companies must perform their own classifications without official guidelines.
Second, the harmonized standards are being developed by CEN-CENELEC JTC 21 and, according to current progress, are not expected until December 2026 at the earliest. The postponement is linked to the availability of these standards. Anyone planning today cannot assume these standards will be available on time.
Third, certain obligations are explicitly excluded from the postponement. This includes, in particular, the transparency obligations under Article 50. Anyone using Joule or another AI-powered assistant in customer-facing contexts must clearly indicate that the user is interacting with an AI starting August 2, 2026. This obligation remains in effect under any scenario.
Practical planning utilizes a dual-track approach:
Track A, basic compliance by August 2026. All obligations that apply independently of the Digital Omnibus will be met on time. This includes, in particular, transparency obligations under Article 50, reporting channels for serious incidents under Article 73, and the inventorying of existing AI systems.
Track B, high-risk readiness by December 2027. The more complex obligations under Article 26 (log retention, conformity assessment, FRIA creation) are planned with the later target date in mind, but are not put on standby. The three-to-six-month lead time required for a conformity assessment remains constant, regardless of the Omnibus.
Decoupling these two tracks prevents a potential final decision against the postponement from causing last-minute panic, and conversely, prevents the need to scrap and restart the entire plan should the postponement be approved.
Political uncertainty. The trilog process could lead to either a watered-down or a strengthened compromise. A delay in reaching an agreement beyond June 2026 would mean that the original deadlines apply, even if a postponement is granted later.
Regulatory arbitrage. This shift creates an incentive to delay projects rather than implement them in compliance. This is operationally risky, as lead times for conformity assessments and FRIA preparation are not getting any shorter.
National deviations. National implementation laws can be interpreted more strictly than the EU requirements. As of April 20, 2026, the German AI Implementation Act is in preparation, but the final text is not yet available.
Do not plan based on the delay. As long as the trilogue is not concluded, the original deadline should be used as the basis for planning. If the delay happens, the schedule becomes a buffer, not a pause button.
Address transparency obligations immediately. Article 50 applies regardless of any ongoing discussions starting August 2, 2026.
Classification without a pause button. Anyone waiting for guidance will have to reclassify under time pressure later. It is better to perform a conservative preliminary classification and adjust it later.
The Digital Omnibus on AI will very likely postpone high-risk obligations until December 2027. However, this delay is no reason to stand down. Transparency obligations under Article 50 remain active as of August 2, 2026, and the Commission's classification guidance is not yet available. Reliable planning requires a dual-track approach that addresses basic compliance immediately and high-risk readiness in stages.