The European Union has agreed a series of targeted adjustments to the implementation framework of the AI Act through a new “Digital Omnibus on AI” package. The agreement, reached between the European Parliament and the Council, aims to simplify compliance obligations while maintaining the core architecture of the AI Act.
Under the agreement, the application dates for several categories of high-risk AI systems will be postponed. Standalone high-risk AI systems are now expected to become applicable from December 2027, while AI systems integrated into regulated products will apply from August 2028.
The package also introduces measures intended to reduce implementation burdens for SMEs and smaller mid-cap companies. These include simplified documentation requirements, clearer interaction between the AI Act and existing sectoral legislation, and broader access to AI regulatory sandboxes designed to support testing and compliance preparation.
The AI Act remains the world’s first comprehensive horizontal regulatory framework for artificial intelligence, establishing obligations based on the level of risk associated with AI applications. High-risk systems used in areas such as employment, healthcare, critical infrastructure, law enforcement, and financial services are subject to stricter requirements relating to transparency, documentation, risk management, and human oversight.
Alongside simplification measures, the Omnibus package also strengthens prohibitions on certain harmful AI uses, including systems designed to generate non-consensual sexually explicit content and child sexual abuse material.
The implementation adjustments follow increasing feedback from businesses, industry associations, and Member States regarding the operational complexity of preparing for AI Act compliance. Concerns had emerged regarding overlapping regulatory requirements, legal uncertainty, and the speed at which generative AI technologies evolved during the legislative process.
For companies across sectors including manufacturing, retail, mobility, healthcare, financial services, and digital technologies, the revised timelines are expected to provide additional time to adapt governance systems, internal controls, supplier management processes, and technical documentation procedures ahead of full implementation.
The agreement also reflects a broader trend within EU policymaking toward balancing regulatory ambition with competitiveness and implementation feasibility. Similar debates have recently emerged around sustainability reporting, due diligence legislation, cybersecurity obligations, and industrial policy frameworks.
The AI Omnibus package now moves toward formal adoption procedures, while the core provisions of the AI Act continue to enter into force progressively over the coming years.