Did the EU Delay the AI Act? What Actually Took Effect on August 2, 2026. Did the EU delay the AI Act? Partly. The Digital Omnibus on AI, given final Council approval on June 29, 2026, deferred high-risk obligations for Annex III systems to December 2, 2027. Article 50 transparency obligations were not deferred and took effect on August 2, 2026, exactly as scheduled.  

The headline most organizations heard was simpler. “The EU delayed the AI Act.” Half true, and imprecise in a way that creates exposure.  

What the Digital Omnibus Actually Changed  

For more than a year, August 2, 2026 anchored enterprise AI compliance planning. It was the date the high-risk regime under Chapter III was scheduled to apply, and it drove budgets and board attention accordingly.  

By late 2025 the deadline was visibly unworkable. Harmonized standards from CEN-CENELEC were running behind, Commission guidelines were still in draft, and many Member States had not resourced their market surveillance authorities. The Commission tabled the Digital Omnibus on November 19, 2025. Parliament endorsed it on June 16, 2026 by 423 votes to 57, and the Council gave final approval on June 29. 

Standalone high-risk systems under Annex III, a category covering hiring tools, credit scoring, and biometric identification, now have until December 2, 2027. High-risk AI embedded in regulated products under Annex I moves to August 2, 2028. Sixteen months of relief on the heaviest part of the regime.  

What Took Effect on August 2 Anyway?  

Article 50 transparency obligations. People must be told when they are interacting with an AI system. Synthetic audio, image, and video content must be marked as artificially generated in machine-readable form. Deepfakes require disclosure, and emotion recognition systems carry their own notice duties.  

None of that moved. Neither did the Article 4 AI literacy obligation.  

One carve-out applies. Article 50(2), which governs provider-side marking of synthetic content, does not reach systems already on the market as of August 2, 2026.  

Penalties under the AI Act run up to 35 million euros or 7 percent of global annual turnover, depending on the provision breached.  

The Next Deadline Is December 2, 2026  

Legacy systems lose the Article 50(2) exemption on that date, and new prohibited practices begin, including an Article 5 ban on AI-generated non-consensual intimate imagery and child sexual abuse material.  

The Omnibus also amended the GDPR, creating a lawful basis to process special category data for bias and fairness testing. Teams previously blocked from testing models for discriminatory outcomes by data protection constraints now have a route, though the change interacts directly with the seven GDPR data protection principles, purpose limitation and data minimization in particular.  

What This Means for Your Compliance Program  

Re-baseline the high-risk roadmap to December 2027 instead of standing down. Grandfathering covers systems placed on the market before the new dates, but only until a substantial modification resets the clock, so map which deployments qualify and define what counts as substantial.  

Organizations using machine learning for document classification should note that the deferral covers Annex III categories, not every AI deployment. Classification depends on use, not technology.  

Messaging Architects helps organizations map AI obligations against their actual data and system inventory rather than their policy language. Read more compliance analysis on our blog, or work with our parent company eMazzanti Technologies to put the technical controls in place. Contact our team to find the gaps.