Update, 7 August 2026: this post originally covered the 2 August 2026 high-risk deadline as originally legislated. That deadline has since been postponed by the Digital Omnibus on AI, adopted by the European Parliament on 16 June 2026 (423 votes to 57) and given final Council approval on 29 June 2026. The rewrite below reflects the current, amended timeline.
The EU AI Act has been in force since August 2024, but its obligations were always designed to phase in over several years. The original text set 2 August 2026 as the date the full high-risk AI system regime, Annex III, became enforceable. That date has moved. Here is exactly what changed, what didn't, and why the extra runway is not a reason to stop preparing.
What the Digital Omnibus actually changed
The Digital Omnibus on AI, first proposed by the European Commission on 19 November 2025, deferred the high-risk compliance deadline. The Council and Parliament settled on two fixed dates:
- 2 December 2027 for stand-alone high-risk AI systems under Annex III, employment, credit, education, health, law enforcement, and the other listed use cases.
- 2 August 2028 for high-risk AI systems embedded in products already regulated under other EU product-safety law (Annex I), for example medical devices or machinery with an AI component.
Publication in the Official Journal was expected in July 2026, with the amended regulation entering into force three days after. In practical terms: the conformity assessment, technical documentation, and EU database registration requirements for Annex III systems are no longer due this August. Companies get roughly sixteen more months.
What still happens on 2 August 2026, unchanged
The Omnibus did not touch everything. These obligations were not part of the delay and take effect, or continue to apply, on schedule:
- Article 50 transparency duties. Chatbot and AI-assistant disclosure, labelling of AI-generated content, and deepfake marking remain live from 2 August 2026.
- GPAI enforcement powers. The Commission's enforcement toolkit over general-purpose AI model providers, in force since August 2025, gains its full penalty regime on this date.
- Prohibited practices. Unchanged, banned since 2 February 2025.
So if your product is a customer-facing chatbot or generates AI content for EU users, the disclosure obligation you already had does not go away. What changes is specifically the Annex III high-risk conformity machinery, not the transparency layer underneath it.
Why "we have until 2027" is the wrong read
It is tempting to treat a 16-month extension as permission to deprioritise. Two reasons that would be a mistake:
- Classification doesn't wait for enforcement. Whether your system is high-risk under Annex III is a fact about what it does, not about the compliance calendar. Knowing your tier now tells you whether you need Article 50 disclosure today, and how much technical documentation work sits ahead of you before late 2027.
- Notified-body capacity is finite. Every company that waited for the original August 2026 deadline, plus every company now targeting December 2027, is competing for the same pool of conformity assessment bodies. A longer runway does not mean a shorter queue closer to the (new) date.
How to use the extra time
- Classify every AI system you build or deploy, now. This costs nothing to do early and tells you which of the items below actually apply to you.
- Confirm your Article 50 disclosure is correct before 2 August 2026. This is the one obligation that did not move.
- Start technical documentation while the system is fresh, rather than reconstructing training data provenance and design decisions from memory in 2027.
- Design human oversight in from the start of any new high-risk feature. Retrofitting it later is more expensive than building it in.
EuroComply's free EU AI Act assessment classifies your system's risk tier in about 3 minutes and returns the exact list of documents your tier requires, with the legal article and current deadline for each, so your team is working from the amended timeline, not the original one.