The European Union has substantially extended the application deadlines under the AI Act (Regulation (EU) 2024/1689, hereinafter: “AI Act”). On the basis of a political agreement reached on 7 May 2026 between the European Commission, the European Parliament and the Council, several application dates have been deferred and a number of substantive adjustments made. The package is informally known as the “AI Omnibus”.
The changes are of immediate practical relevance for businesses that develop, deploy or distribute AI systems. The following article sets out the key amendments and their implications for compliance planning.
1. Background: The AI Act and Its Staggered Application Timetable
The AI Act entered into force on 1 August 2024. Its provisions apply, as a general rule, from 2 August 2026, although the Regulation itself provides for differentiated application periods: the prohibitions on certain AI practices under Chapter II have already applied since 2 February 2025. The extensive obligations for providers and deployers of high-risk AI systems listed in Annex III were originally intended to apply from 2 August 2026.
The AI Omnibus reflects the legislature’s assessment that businesses and public bodies require additional lead time to implement the Regulation’s often complex requirements. In addition, the package introduces targeted substantive corrections, in particular with regard to the scope of application for machinery.
2. Transparency and Labelling Obligations under Art. 50 AI Act: A Differentiated Transitional Arrangement
Article 50 of the AI Act imposes transparency obligations on AI systems and AI-generated content. For AI-generated content – that is, automatically produced texts, images, audio and video material – Art. 50(2) AI Act provides for a labelling obligation. That obligation is deferred by the AI Omnibus to 2 December 2026.
To be distinguished from this are the obligations under Art. 50(4) AI Act. These apply to AI systems that synthetically generate or manipulate images, audio or video content (the so-called deepfake labelling requirement). These obligations are unaffected by the AI Omnibus and continue to apply from 2 August 2026.
In practice, this means: providers and deployers using AI systems to generate texts, images or other content have until December 2026 to align their processes and systems with the labelling requirements of Art. 50(2) AI Act. Those operating AI systems that generate deepfake content within the meaning of Art. 50(4) AI Act must, however, comply with the relevant obligations as of August 2026.
3. High-Risk AI under Annex III: Extension of the Deadline by Approximately 16 Months
Annex III of the AI Act lists AI systems classified as high-risk on account of their area of application. These include, inter alia, systems in the fields of employment and human resources management, education and vocational training, access to essential private and public services, and law enforcement and border control. The applicable obligations for providers and deployers of such systems – ranging from technical documentation and risk management systems through to logging requirements and conformity assessment procedures – will, under the AI Omnibus, no longer apply from 2 August 2026, but only from 2 December 2027.
This extension of approximately 16 months affords affected businesses and public authorities considerably more time to establish the requisite compliance structures. That said, the additional time should not be treated as grounds for inaction: the requirements for high-risk AI systems are extensive and presuppose structured preparatory work that should commence without delay.
4. High-Risk AI Embedded in Products under Annex I
Annex I of the AI Act refers to a number of product safety frameworks, including the Machinery Regulation, the Medical Device Regulation, the Toy Safety Directive, and the Type Approval Regulation for motor vehicles. AI systems embedded in products subject to those legislative acts are governed by their own transitional provisions.
Under the AI Omnibus, an even later application date applies to these systems: the relevant high-risk obligations for AI embedded in Annex I products will not apply until 2 August 2028. Manufacturers integrating AI functions into regulated products thus benefit from a substantially extended transition period and are able to plan their product architectures and certification strategies on a long-term basis.
5. Machinery in Annex III: Reclassification from Section A to Section B
A particularly significant substantive change concerns the scope of the AI Act as it applies to machinery. Certain machinery was previously listed in Section A of Annex III. The AI Omnibus reclassifies that machinery into Section B of the same Annex.
This reclassification has far-reaching consequences. Under Art. 2(2) AI Act, AI systems listed in Section B of Annex III that are embedded in products subject to the product safety legislation referred to in Annex I are subject to a substantially reduced set of obligations: a substantial portion of the requirements applicable to high-risk AI systems no longer applies. Manufacturers of AI-enabled machinery should therefore carefully assess whether their products now fall within this exemption and identify which obligations are consequently no longer applicable.
6. AI Regulatory Sandboxes: Extension of the Establishment Deadline
The AI Act obliges the competent national authorities to establish so-called AI regulatory sandboxes. Within these environments, companies may develop and test innovative AI systems under regulatory supervision and under relaxed conditions, prior to their market placement. The original deadline for the establishment of these sandboxes by the competent authorities – 2 August 2026 – is extended by the AI Omnibus to 2 August 2027.
For businesses that depend on access to regulatory sandboxes – such as AI start-ups or companies with novel AI applications – this means that this facility will not be available on a broad basis before mid-2027. This should be factored into development and market launch planning.
Practical Note
The AI Omnibus affords affected businesses an important breathing space – it does not, however, relieve them of the obligation to prepare. The additional time should be used to conduct a structured stocktake: which AI systems are being deployed or developed within the organisation? Which category of the AI Act applies? And might products incorporating embedded AI benefit from the exemption for machinery under Section B? The answers to these questions should be developed now – not in the period immediately before the new deadlines expire.
Recommended Actions
- AI system inventory: Systematically record all AI systems deployed, developed or distributed within your organisation and assign them to the relevant categories under the AI Act. Particular attention should be paid to the distinction between high-risk AI under Annex III and Annex I.
- Review of high-risk classification: For systems with a potential connection to Annex III, a legal assessment should be conducted. In particular, it should be examined whether the reclassification of machinery into Section B of Annex III permits a reassessment and results in obligations no longer being applicable.
- Building AI governance structures: Use the extended deadlines to assign internal responsibilities, establish documentation systems, and plan training for relevant members of staff.
- Monitoring regulatory developments: The AI Omnibus demonstrates that the regulatory landscape remains in flux. Continuous monitoring at EU level and flexible compliance planning are indispensable.
Contact:
Jens Borchardt