Digital Omnibus 2026: What has changed in the EU AI Act

What the Digital Omnibus is and since when it has applied

Since July 27, 2026, Regulation (EU) 2026/1744 has applied. It was adopted on July 8, 2026, published in the Official Journal on July 24, 2026, and amends the EU AI Act (Regulation (EU) 2024/1689) in several places. What was discussed for months as a project under the short name "Digital Omnibus on AI" is thus applicable Union law.

The legal act extends deadlines, removes individual powers, and reformulates a central obligation. It neither repeals the AI Act nor postpones it as a whole.

Limit: Which provisions are affected in detail follows solely from the amending text. Sweeping statements about "the postponed AI Act" are incorrect — the postponements are narrowly limited.

These deadlines have been postponed by the Digital Omnibus

Above all, the obligations for high-risk AI were postponed — by 16 and 12 months respectively. The overview shows the three postponed dates of application:

• High-risk AI under Annex III (Art. 6 (2) EU AI Act) — Previous date of application: August 2, 2026. New date of application: December 2, 2027.
• High-risk AI in products under Annex I (Art. 6 (1) EU AI Act) — Previous date of application: August 2, 2027. New date of application: August 2, 2028.
• Establishment of national AI regulatory sandboxes (Art. 57 EU AI Act) — Previous date of application: August 2, 2026. New date of application: August 2, 2027.

The new dates are set out in the amended Article 113 of the EU AI Act; recital 40 of the amending regulation gives the reasons for them. They are confirmed by the Commission on its AI Act page (as of 3 Aug 2026). The deadline for the regulatory sandboxes is given by the Council of the EU in its communication of May 7, 2026.

A fourth, narrowly framed amendment concerns Article 50: for systems placed on the market before August 2, 2026, the machine-readable marking of synthetic content under Article 50 (2) only applies from December 2, 2026. It covers neither the chatbot notice under paragraph 1 nor the deepfake labeling under paragraph 4 — both apply without a grace period.

The omnibus does not only bring relief: Article 5 is supplemented by a new prohibition: AI systems that generate or manipulate non-consensual intimate material or depictions of child sexual abuse. It applies from December 2, 2026 (European Commission, 27 Jul 2026; Art. 5 in conjunction with Art. 113 EU AI Act as amended by Regulation (EU) 2026/1744, as of 5 Aug 2026).

Limit: A later date of application is not an exemption. Anyone who operates a system that will later fall under Annex III gains preparation time — nothing more.

Article 4 EU AI Act: the obligation remains, the standard is lowered

Article 4 was replaced by Article 1 point 5 of the amending regulation. Paragraph 1 now reads:

"Providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf […]. This obligation shall not oblige providers or deployers to guarantee a specific level of AI literacy for any person."

The previous version required them "to ensure, to their best extent," that the persons concerned have "a sufficient level of AI literacy". This reference to outcomes — already limited to best efforts there — has been dropped. What remains is the obligation itself: it has applied unchanged since February 2, 2025, and it is addressed to providers and deployers. Two paragraphs have been added: under paragraph 2, the Commission and the Member States support implementation and publish practical examples on a central information platform; under paragraph 3, the AI Board adopts recommendations on this. The Commission explains this in its questions and answers on AI literacy (as of 27 Jul 2026).

In practice, this means: an obligation to establish a level of competence has become an obligation to take measures to promote competence. What belongs in such a measure is described in the article on the content of AI literacy training.

Limit: The training obligation has not been "abolished". Anyone who concludes from the new version that there is nothing left to do reads the wording against its own text.

What has been enforceable since August 2, 2026

The EU AI Act has in principle been applicable since August 2, 2026 — with exceptions; the Commission announced the start of enforcement on July 31, 2026. Three things above all took effect: the transparency obligations of Article 50, the enforcement activity of the AI Office together with the national authorities, and, in the area of penalties, Article 101.

In practice, Article 50 is the most noticeable. Among other things, it requires that people can recognize when they are talking to an AI system and that deepfakes are labeled. How this looks for a service chatbot is explained in the article on the labeling obligation under Article 50.

The obligations for general-purpose AI models, on the other hand, are not new: they have already applied since August 2, 2025. Anyone who presents them as a 2026 innovation is off by twelve months. An overview of the system behind this is provided in the article on the four AI risk classes.

Limit: Applicability and actual enforcement are two different things. The fact that an authority is competent says nothing about how it prioritizes.

Who enforces the AI Act in Germany

In Germany, the Bundesnetzagentur has been competent since July 29, 2026. The legal basis is the Act implementing Regulation (EU) 2024/1689, signed on July 22, 2026, and promulgated on July 28, 2026 (BGBl. 2026 I no. 223); its Article 1 is the KI-MIG.

According to the communication from the Federal Ministry for Digital Transformation and Government Modernisation of July 29, 2026, the Bundesnetzagentur becomes the "market surveillance authority, point of contact, and complaints office for the AI Act". The Act bundles coordination there in a coordination and competence center (KoKIVO); in addition, the authority operates an AI Service Desk as a low-threshold information service, in particular for small and medium-sized enterprises.

Limit: The Bundesnetzagentur is not the sole supervisory authority. The Act expressly builds on the existing sectoral market surveillance authorities; in the media sector, the federal states make the designation.

What this means for ongoing implementation projects

The time pressure has shifted, not dissolved. Anyone who has set up an Annex III project has breathing room until December 2027. Anyone who uses chatbots, assistance systems, or generative tools, on the other hand, has been subject to applicable law since August 2, 2026 — and the literacy obligation of Article 4 has been running since February 2025 anyway.

A practical note on working with sources: in early August 2026, the Commission's AI Act Service Desk still reflected the status before the omnibus for several article texts and in its implementation timeline — for Article 113 and Article 50 with an explicit note on the amendment, for Article 4 and Article 57 without. Anyone who takes deadlines from there takes outdated dates.

Limit: This article describes the legal position, not how a specific system is affected. Whether a particular AI system falls under Annex III is a question for the individual case.

Has the EU AI Act been abolished by the Digital Omnibus?

No. Regulation (EU) 2026/1744 amends the EU AI Act; it does not repeal it. Above all, the dates of application for high-risk AI were postponed. The transparency obligations of Article 50 and the literacy obligation of Article 4 continue to apply unchanged.

Do the high-risk obligations now apply from August 2026 or not?

No longer. Since the amendment, December 2, 2027, applies to stand-alone high-risk systems under Annex III and August 2, 2028, to product-embedded systems under Annex I. Texts that still state August 2, 2026, reflect an outdated status.

Does our company still have to offer AI training?

Article 4 EU AI Act continues to oblige providers and deployers to take measures to support the development of AI literacy of their staff. What is new is the explicit clarification that no specific level of competence of individual persons has to be guaranteed. The obligation has existed since February 2, 2025.

Which authority is competent in Germany?

Since July 29, 2026, the KI-MIG has assigned market surveillance to the Bundesnetzagentur, which is also the point of contact and complaints office. Existing sectoral market surveillance authorities remain competent alongside it; in the media sector, the federal states make the designation.

What specifically became applicable on August 2, 2026?

The EU AI Act became applicable in principle. Above all, the transparency obligations of Article 50 and the enforcement activity of the AI Office together with the national authorities took effect. The obligations for general-purpose AI models, on the other hand, have already applied since August 2, 2025.

• Regulation (EU) 2026/1744 of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 as regards the simplification of the implementation of harmonised rules on artificial intelligence ("Digital Omnibus on AI") – OJ L, 2026/1744, 24 Jul 2026, in force 27 Jul 2026 (accessed 5 Aug 2026)
• Regulation (EU) 2024/1689 (EU AI Act), original version OJ L, 12 Jul 2024 – Art. 4, Art. 5, Art. 6, Art. 50, Art. 57, Art. 101, Art. 113 (accessed 5 Aug 2026)
• Regulation (EU) 2024/1689, consolidated version (EUR-Lex consolidation, amendment status after Regulation (EU) 2026/1744) (accessed 5 Aug 2026)
• European Commission, "AI Omnibus enters into force", 27 Jul 2026 (accessed 5 Aug 2026)
• European Commission, "Commission starts enforcing AI Act rules and new transparency requirements 2 August", 31 Jul 2026 (accessed 5 Aug 2026)
• European Commission, policy page "AI Act" – as of 3 Aug 2026 (accessed 5 Aug 2026)
• European Commission, "AI Literacy – Questions & Answers" – as of 27 Jul 2026 (accessed 5 Aug 2026)
• Council of the European Union, press release of 7 May 2026 (accessed 5 Aug 2026)
• Act implementing Regulation (EU) 2024/1689 (Art. 1: KI-MIG) – BGBl. 2026 I no. 223, signed 22 Jul 2026, promulgated 28 Jul 2026 (accessed 5 Aug 2026)
• Federal Ministry for Digital Transformation and Government Modernisation, „Neues KI-Gesetz tritt in Kraft“, 29 Jul 2026 (accessed 5 Aug 2026)
• Bundesnetzagentur, AI Service Desk (accessed 5 Aug 2026)