Is the EU AI Act delayed? Yes — but not the part that probably applies to you
If you've seen headlines saying the EU AI Act has been delayed, they're half right — and the half that's wrong is the half that matters for most companies. Here is the position as of July 2026, in plain language.
What actually happened
In November 2025 the European Commission proposed a "Digital Omnibus" — a package simplifying several EU digital laws, including targeted amendments to the AI Act. After months of negotiation, the European Parliament adopted the agreement on 16 June 2026 and the Council gave its final approval on 29 June. Publication in the Official Journal is expected in July 2026, with the amendments entering into force three days later.
The Omnibus delays the AI Act's high-risk obligations. It does not delay the Act's transparency obligations. Those two categories affect very different companies — and far more organisations sit in the second category than realise it.
What moved, what didn't
| Obligation | Status | Date |
|---|---|---|
| High-risk AI systems — standalone (Annex III: recruitment, credit scoring, education, etc.) | Deferred | 2 Dec 2027 |
| High-risk AI embedded in regulated products (Annex I) | Deferred | 2 Aug 2028 |
| Telling users they're interacting with an AI system (chatbots, voice agents) | Not delayed | 2 Aug 2026 |
| Disclosing deepfakes and AI-generated media | Not delayed | 2 Aug 2026 |
| Disclosing AI-generated text published to inform the public | Not delayed | 2 Aug 2026 |
| Notifying people exposed to emotion recognition or biometric categorisation | Not delayed | 2 Aug 2026 |
| Machine-readable watermarking of AI-generated content — systems already on the market | Short deferral | 2 Dec 2026 |
| Watermarking — systems launched from 2 Aug 2026 | From launch day | No grace period |
| New prohibition: AI systems generating non-consensual intimate imagery without safeguards | New (added by Omnibus) | 2 Dec 2026 |
Why 2 August 2026 is a real deadline
Three reasons. First, the Article 50 transparency duties apply from that date on the original schedule — the Omnibus text confirms rather than moves them. Second, national market surveillance authorities receive their full investigatory and enforcement powers on the same date, formally ending the guidance-first phase. Third, the fine ceiling for transparency violations is up to €15 million or 3% of global annual turnover — not the Act's top tier, but far from symbolic.
The practical risk in year one is less about regulators hunting for missing chatbot banners and more about complaints. A competitor, a disgruntled user, or a journalist can put your missing disclosure in front of a regulator that now has the power — and the obligation — to look into it. What separates a warning from an open file is usually evidence: an inventory of your AI systems, a record of how you scoped each one, and dated proof of the disclosures you deployed.
What to do this month
The work is smaller than it sounds for most companies: list every AI system you provide or deploy (including procured tools); run each through the five Article 50 questions — does it talk to people, generate content, publish synthetic media, publish AI-written text, or analyse emotions/biometrics; deploy the matching disclosure wording for every "yes"; and keep dated evidence. If you ship generative features, hand engineering the watermarking requirement with a 2 December delivery date now, not in November.
The working documents, ready to deploy
Our EU AI Act Transparency Compliance Pack contains the scoping decision tree, all five disclosure templates, an auto-computing AI inventory register, the watermarking checklist, and a pre-written leadership briefing — plain language, adaptable in an afternoon, with free updates through 2 December 2026.
Get the pack — €79 Free deadline map