EU AI Act · Digital Omnibus

What changed on 2 August 2026 — and what still applies

The Digital Omnibus moved several AI Act deadlines. “The AI Act is delayed” is only half true — here is precisely what shifted, and what you are still on the hook for.

Aug 2, 2026
Art. 50 transparency — unchanged
Dec 2026
New prohibitions
Dec 2027
High-risk (Annex III) postponed
By · Editorial Team
Published Reviewed
Last verified · Jul 14, 2026 Last checked · 14 Jul 2026, 09:00 UTCSourced from official primary sources — digital-strategy.ec.europa.eu ↗.
In one minute
  • still appliesArticle 50 transparency still starts 2 August 2026. If you run a chatbot, generate or publish AI/deepfake content, or use emotion-recognition or biometric-categorisation, the disclosure and labelling duties apply on that date. This did not move.
  • postponedHigh-risk (Annex III) obligations were postponed to December 2027. The heavier conformity, risk-management and documentation duties for Annex III high-risk systems got more runway.
  • newA further set of prohibitions applies from December 2026. Some banned practices phase in on the Omnibus schedule — check whether your use case is caught.
The timeline

Every deadline, in order

  1. 2 Feb 2025in force

    First prohibitions (Art. 5 unacceptable-risk practices) + AI-literacy duties took effect.

  2. 2 Aug 2025in force

    General-purpose AI (GPAI) model rules + the governance / notified-body framework took effect.

  3. 2 Aug 2026still applies

    Article 50 TRANSPARENCY duties apply — disclose AI chatbots, label AI-generated / deepfake content, and inform people of emotion-recognition & biometric-categorisation use. This did NOT move.

  4. Dec 2026new

    A further set of PROHIBITIONS becomes applicable under the Digital Omnibus schedule.

  5. Dec 2027postponed

    High-risk (Annex III) obligations — POSTPONED here from 2 Aug 2026 by the Digital Omnibus.

still applies

Due 2 August 2026

Article 50 is a transparency article, not a high-risk one — so the Annex III postponement did not touch it. In scope, you must broadly:

  • tell people when they are interacting with an AI system (e.g. a chatbot), unless it is obvious;
  • mark AI-generated or manipulated audio, image, video or text (incl. deepfakes) machine-readably;
  • disclose emotion-recognition and biometric-categorisation use to the people exposed.
postponed

Pushed to December 2027

The Annex III high-risk obligations got more runway:

  • risk management & data governance;
  • technical documentation & logging;
  • human oversight & conformity assessment.

The direction of travel is unchanged — build the compliance file now, not at the deadline.

Go deeper — one-time report

Omnibus Impact Report

Omnibus Impact Report = the current EU Digital Omnibus delta + a company-specific applicability map

A superset of the standard AI Compliance Report: every deadline the EU Digital Omnibus moved — Aug 2 2026 transparency, Dec 2026 prohibitions, Dec 2027 high-risk — PLUS a company-specific applicability map showing which of those obligations actually hit your business. Roughly twice the standard report.

  • every date on this page, mapped to the exact obligations it moves;
  • a company-specific applicability map — which of those obligations actually hit your business;
  • an Article 50 vs Annex III “what you owe now vs later” split;
  • source citations for each change — audit-ready, not a summary.

A superset of the standard AI Compliance Report ($79) — roughly twice the depth.

$149
one-time · no subscription

Track every one of these deadlines automatically

The AI Law Tracker API returns EU + US + global AI-law records — status, effective and enforcement dates, and official sources — as clean JSON. Wire the Article 50 date (and every other one on this page) straight into your compliance dashboard. Free key to start.

Get a free API key →Browse EU AI Act records
Verify it yourself

Primary sources