EU Delays Most of the AI Act's High-Risk Rules

European Union lawmakers reached a provisional agreement on May 7 to slim down and stagger parts of the EU AI Act, the bloc's flagship law on artificial intelligence. Most of the headline obligations for high-risk AI systems will not take effect until late 2027 or 2028.

The deal, known as the AI Omnibus, was negotiated under tight schedule between the Council of the EU and the European Parliament. Its central feature is a delay. For standalone high-risk systems classified under Annex III of the Act, including those used in employment, education, credit scoring and biometric identification, the application of the requirements is deferred to December 2, 2027. For high-risk AI embedded in regulated products under Annex I, including medical devices and machinery, the deadline moves to August 2, 2028.

Watermarking obligations under Article 50(2), which apply to any AI system that generates synthetic content, are pushed back to December 2, 2026. The original deadline was August 2026.

The Omnibus also introduces a new prohibition. AI systems specifically designed to generate non-consensual intimate imagery and child sexual abuse material, including the so-called "nudifier" tools that have proliferated on consumer app stores, will join the list of practices banned outright under the Act.

The political read is that Brussels is buying time. Industry groups have argued for months that compliance was impossible without supporting technical standards, and the Commission's Digital Omnibus initiative was launched in late 2025 with simplification as an explicit goal. Civil society groups counter that delay weakens enforcement. Tech Policy Press notes that the deferral is also a signal to Washington and London about the pace at which strict rules can be made to bite.

The agreement remains provisional. It must still be formally endorsed and adopted by the co-legislators before becoming law. The shift comes against a busy regulatory backdrop: Connecticut recently sent its own sweeping AI bill to the governor's desk, and the White House is weighing pre-release safety reviews in the United States.

Detailed implementation summaries are available from Hogan Lovells, IAPP, and Modulos.