EU AI Act After the Digital Omnibus: What Still Applies on 2 August 2026

The Digital Omnibus moved the EU AI Act’s high-risk deadlines by more than a year. It did not move most of Article 50. This analysis separates what changed, what still applies on 2 August 2026, and what organizations should verify now.

This is not an argument that the delay is unimportant. It is an argument for reading the amendment precisely: the postponement covers one regime, not the Act, and the obligations that remain scheduled are the ones most organizations have examined least.

What the Digital Omnibus changed

The European Parliament gave its final approval on 16 June 2026 and the Council adopted the measure on 29 June 2026. Regulation (EU) 2026/1744, the Digital Omnibus on AI, was published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026. The amended application dates for Chapter III Sections 1, 2 and 3, with the stated exception in Article 113, are 2 December 2027 for AI systems classified as high risk under Article 6(2) and Annex III and 2 August 2028 for systems classified under Article 6(1) and Annex I. The exact scope, including the interaction with relevant sectoral product legislation, must be assessed for each system.

The stated reason for the deferral was readiness: national authorities and harmonized technical standards were not in place. The requirements themselves, including conformity assessment, technical documentation and human oversight controls, remain scheduled.

What still applies on 2 August 2026

Article 50 applies from 2 August 2026. It is not a single blanket duty. The relevant obligation depends on the actor’s role, the system’s intended use, the type of output, the audience and the conditions and exceptions in the applicable paragraph. A targeted transition applies to Article 50(2) for certain existing generative systems, as explained below.

• AI interaction disclosure — Article 50(1): providers of AI systems intended to interact directly with natural persons must design and develop them so the persons concerned are informed that they are interacting with AI, unless this is obvious from the perspective and circumstances specified in the Act. A specific law enforcement exception applies subject to the statutory conditions.
• Synthetic content marking — Article 50(2): providers of AI systems, including general purpose AI systems, generating synthetic audio, image, video or text content must ensure that outputs are marked in a machine readable format and detectable as artificially generated or manipulated. The technical solution must be effective, interoperable, robust and reliable as far as technically feasible, taking account of the factors stated in the Act. The paragraph includes specific exceptions, including qualifying standard editing or non-substantial alteration and certain legally authorised criminal-law uses.
• Emotion recognition and biometric categorisation — Article 50(3): deployers must inform natural persons exposed to the operation of these systems and process personal data under the applicable data protection framework. The paragraph contains a specific law enforcement exception and conditions.
• Deepfake disclosure — Article 50(4): deployers of a system generating or manipulating image, audio or video content constituting a deepfake must disclose that it has been artificially generated or manipulated, subject to the statutory exception. Evidently artistic, creative, satirical, fictional or analogous works have tailored disclosure treatment.
• Public interest text — Article 50(4): deployers of a system generating or manipulating text published to inform the public on matters of public interest must disclose that it is artificially generated or manipulated. The exemption requires a process of human review or editorial control and a natural or legal person holding editorial responsibility for publication; the paragraph also contains a specific law enforcement exception. Article 50(5) requires the relevant information to be clear, distinguishable, accessible and provided no later than the first interaction or exposure.

A narrow legacy transition

Providers of AI systems, including general purpose AI systems, generating synthetic audio, image, video or text content that were placed on the market before 2 August 2026 must take the necessary steps to comply with Article 50(2) by 2 December 2026. This is a targeted transition for that provider-side marking and detectability obligation. It is not a general postponement of Article 50. The Commission also states that content generated before 2 August 2026 need not be labelled retroactively, while encouraging relevant deployers to do so where possible.

Non-compliance with Article 50 falls within the administrative-fine tier in Article 99(4): up to EUR 15 million or, for an undertaking, up to 3 percent of total worldwide annual turnover for the preceding financial year, subject to the applicable statutory rules and individual circumstances. The lower-of rule applies to SMEs including start-ups and the amendment extends a lower-of rule for the relevant Article 99(4) and (5) fines to small mid-cap enterprises. This is a maximum framework, not a prediction of a fine in a particular case. Territorial scope, actor role, applicable exception, competent authority and national enforcement rules require case-specific legal assessment.

NOMAMIND interpretation: the sixteen additional months are runway, not relief. Transparency duties are pathway duties: disclosure at the first interaction, marking that survives real publication workflows, and role clarity per system are properties of how a decision pathway is designed, not labels added at the end. Systems keep operating through the gap years, and drift does not wait for enforcement dates. Organizations that use the runway to build accountable oversight may be better positioned for the applicable high-risk date. Those that treat a deferral as cancellation may discover gaps only after a consequential decision or supervisory review. This is NOMAMIND interpretation, not a legal conclusion or guarantee of compliance.

What organizations can do now

1. Map roles and EU touchpoints system by system: provider or deployer, and which systems interact with people in the Union.
2. Treat the five Article 50 duties separately; they bind different actors with different exceptions.
3. Identify legacy systems that genuinely qualify for the December 2026 marking transition, and nothing else.
4. Test disclosures and markings in the live workflow: at first interaction, surviving distribution, accessible.
5. Document every exception relied on, including obviousness and editorial responsibility.
6. Keep the relevant high-risk work moving against the applicable classification and date: 2 December 2027 for Article 6(2)/Annex III systems and 2 August 2028 for Article 6(1)/Annex I systems, subject to the amended Act and relevant sectoral legislation.

Primary sources

• European Parliament press release, 16 June 2026
• Council of the EU press release, 29 June 2026
• Article 50, EU AI Act (Commission AI Act Service Desk)
• Commission-facilitated Code of Practice on AI-generated content
• Regulation (EU) 2026/1744, Official Journal
• Commission Guidelines on Article 50
• Commission Article 50 FAQ

Updated 30 July 2026: publication and entry into force of Regulation (EU) 2026/1744 confirmed; Article 50 roles, transition and exceptions clarified.