The EU AI Act Just Changed. Standing Down Would Be a Mistake.
The Digital Omnibus on AI entered into force on 27 July 2026 and pushed the high-risk regime out by sixteen months — but left untouched the obligations most likely to touch your business, which land on 2 August 2026.
For two years, 2 August 2026 was treated as a single monolithic deadline. That framing is now wrong, and the way in which it is wrong matters commercially. Regulation (EU) 2026/1744 — the Digital Omnibus on AI — entered into force on 27 July 2026. It defers the obligations furthest from being implementable, and leaves untouched the obligations already ready to go.
The revised compliance calendar
- 2 Aug 2025 — GPAI model obligations already in force.
- 2 Aug 2026 — Article 50 transparency obligations: chatbot disclosure, deepfake labelling, emotion-recognition notices.
- 2 Dec 2026 — Machine-readable marking for generative systems already on market; new Article 5 prohibitions.
- 2 Dec 2027 — High-risk obligations for stand-alone Annex III systems (employment, credit, education). Deferred from 2 Aug 2026.
- 2 Aug 2028 — High-risk obligations for AI embedded in Annex I regulated products.
Why the extension is worth less than it looks
A credible high-risk conformity programme is a six to twelve month exercise per system. Sixteen months of relief is one preparation cycle plus contingency — not a reprieve. The penalties did not change: up to €35M or 7% of global turnover for prohibited practices. And your enterprise customers are not waiting for the regulator — they have already absorbed AI Act vocabulary into their vendor questionnaires.
The efficient path
The most cost-effective route is not a bespoke EU-compliance programme. It is an ISO/IEC 42001 AIMS with the Act as a regulatory overlay. In our assessments, a certified AIMS delivers over 80% coverage of high-risk obligations — the remaining delta being EU-specific registration and conformity assessment. The sixteen months you just gained has an obvious use: build the management system now.