EU AI Act Digital Omnibus Explained

The EU AI Act's most-cited deadline, 2 August 2026 for high-risk system obligations, no longer applies. In mid-2026, EU legislators agreed a legislative package known as the Digital Omnibus that pushed it back by 16 months, along with a separate delay to the Act's transparency rules.

What actually changed

The European Commission published its original Digital Omnibus proposal on 19 November 2025, aiming to simplify and defer parts of the AI Act's implementation timeline. After negotiation between the Parliament and Council, the final package was formally adopted, with Parliament's endorsement on 16 June 2026 and the Council's final green light on 29 June 2026. It entered into force shortly after publication in the EU's Official Journal, in early July 2026.

Two specific dates moved. High-risk AI system obligations for standalone Annex III systems, hiring and worker-management tools, credit scoring, insurance pricing, and similar, now apply from 2 December 2027, not 2 August 2026. High-risk AI embedded in products already covered by separate safety legislation, like medical devices or machinery, moved to 2 August 2028. Separately, the Article 50 transparency obligations, telling a customer they're talking to an AI chatbot, labelling AI-generated content designed to look real, now apply from 2 December 2026, not 2 August 2026.

Why the delay happened

The owner of a 25-person EU-based HR tech company had already started building conformity assessment processes ahead of the original 2 August 2026 date, treating it as fixed and immovable. The Commission's own stated rationale for the Omnibus was different: giving businesses, standards bodies, and notified bodies more realistic lead time to actually implement the high-risk requirements properly, rather than rushing a compliance regime that technical standards and guidance hadn't fully caught up with yet. It's a delay aimed at implementation quality, not a policy reversal on what the Act ultimately requires.

What to do if you'd already started preparing

Nothing about the underlying obligations changed, only the timing. If you'd started a conformity assessment, built a risk management system, or begun EU AI database registration work, none of that effort is wasted, it's simply no longer needed by a specific date in the next few weeks. The practical move is to recalibrate the internal deadline, not scrap the project. A compliance sprint originally scheduled for July 2026 can reasonably become a longer-lead project targeting late 2027, with the extra runway used to get the work done properly rather than under pressure.

If you hadn't started yet because the original date felt too close to bother with, that reasoning no longer applies either. 2 December 2027 is still a real, binding date for any business using a genuinely high-risk AI system, not an indefinite postponement. See our guide on the full EU AI Act compliance deadlines for the complete, current timeline.

Does this affect anything else in the Act?

The Digital Omnibus's changes were specifically targeted at the high-risk and transparency timelines. The dates that had already passed before mid-2026, prohibited practices from 2 February 2025 and general-purpose AI model obligations from 2 August 2025, are unaffected and remain in force. The AI literacy obligation under Article 4, also already binding since February 2025, is likewise untouched by this package.

What this means for the AI tools you already use

Most SMBs interact with the EU AI Act as a deployer, a business using someone else's AI tool, rather than as a provider building one. For a deployer, the delay mainly changes when a high-risk tool's vendor needs to have finished their own conformity assessment and registration, not what the deployer themselves has to do day to day. If you use an ordinary AI writing, scheduling, or customer-service tool, none of this timeline change affects you, because those tools were never high-risk in the first place. If you use a genuinely high-risk tool, hiring software, credit scoring, insurance pricing, it's worth asking the vendor directly whether their own compliance timeline has shifted in response to the Omnibus, since a vendor working to the old date might finish early, and one working to the new date has more runway than they may have originally communicated.

The transparency obligations move affects a slightly different group: any business running its own AI chatbot or generating AI content customers might mistake for human-made. If you'd already added a disclosure line to your chatbot ahead of the original August 2026 date, keep it, there's no requirement to remove early compliance. If you hadn't gotten to it yet, the pressure is lower than it was, but 2 December 2026 is still a real date worth having on a calendar rather than treated as indefinitely deferred.

Related reading: our AI governance by region.