EU negotiators agree targeted revisions to AI Act implementation timelines and obligations

Negotiators from the European Parliament and the Council presidency have reached a provisional agreement on a set of targeted changes to the EU’s Artificial Intelligence Act.

The revisions are part of the Omnibus VII package, which aims to simplify implementation of existing EU digital legislation while reducing administrative burdens for companies.

One of the main changes concerns implementation timelines for high-risk AI systems. Under the agreement, stand-alone high-risk systems would become applicable from December 2027, while high-risk systems embedded in products would follow in August 2028.

The deal also adjusts transparency obligations. Requirements related to AI-generated content would apply earlier than initially proposed, with the grace period shortened from six months to three months.

Another notable addition concerns prohibited practices. Negotiators agreed to explicitly ban AI systems used to generate non-consensual intimate content and child sexual abuse material.

The agreement also revises the division of oversight responsibilities between the EU AI Office and national regulators. This is particularly relevant for providers developing both general-purpose AI models and downstream systems, an area where supervisory authority had remained unclear.

Sector-specific issues were also addressed. The text introduces mechanisms to avoid overlap between the AI Act and existing product legislation, including for medical devices and machinery. In some sectors, AI-specific requirements will instead be handled through delegated or implementing acts.

The changes reflect a broader effort within the EU to make the implementation of digital regulation more operationally manageable without reopening the core structure of the AI Act.

The agreement is still provisional and must be formally approved by both the European Parliament and the Council before entering into force.