Digital Omnibus on AI: update

Authors: Markus Gaderer and Ilka Kuci

Things are moving forward: On June 16th, 2026, the European Parliament adopted the proposal for the Digital Omnibus Regulation on AI in its amended version, thereby aligning itself with the provisional agreement reached in the trilogue with the Council on May 7th, 2026. What remains is the formal adoption by the Council.

Here is an overview of the key changes:

New deadlines for high-risk systems. Standalone high-risk AI systems will become subject to their obligations starting December 2nd, 2027. However, if they are embedded as safety components in products and fall under sector-specific EU regulations on safety and market surveillance, the applicable date is August 2nd, 2028.
A clear stance against deepfake abuse. A new provision has been added banning certain AI systems from generating non-consensual intimate or sexual content, as well as depictions of child sexual abuse, targeting primarily systems capable of generating such image, video, or audio content. Affected providers and operators have until December 2nd, 2026, to adapt their systems.
More flexibility on labeling. Certain labeling obligations for AI-generated content will also take effect later. For systems placed on the market before August 2nd, 2026, the obligation to provide machine-readable labeling of artificially generated or manipulated content will only apply from December 2nd, 2026.
Registration – a reversal compared to the Commission’s proposal. Where the Commission had originally proposed to remove certain registration obligations, the agreement now provides for a simplified registration requirement. This applies to AI systems for which providers assume that they are exempt from classification as a high-risk AI system.
Bias detection remains in place. The envisaged Article 4a of the AI Act on the processing of special categories of personal data for the detection and correction of bias is retained. The applicable standard is strict necessity, tied to appropriate safeguards.
Who is responsible? The role of the Artificial Intelligence Office is clarified: It is to supervise certain AI systems based on general-purpose AI models. At the same time, cases are delineated in which national authorities remain at the helm – such as law enforcement, border management, judiciary, and financial institutions.
Tensions with product safety law – defused. Finally, the relationship between the AI Act and sectoral product safety regulations is addressed. For specific product areas – such as medical devices, toys, elevators, machinery, and watercraft – overlaps between the AI Act and sectoral regulations are to be avoided. A special regime is foreseen for the Machinery Regulation: Certain provisions of the AI Act will not apply directly; instead, the Commission will specify requirements relating to high-risk AI systems via delegated acts to the Machinery Regulation.