EU AI Act Receives First Major Amendment Since Launch

On May 28-31, 2026, negotiators from the Council of the European Union, the European Parliament, and the European Commission reached a provisional agreement on the Digital Omnibus on AI—the first significant amendments to the EU AI Act since its groundbreaking adoption in June 2024. The agreement brings relief to developers grappling with strict compliance timelines while introducing new prohibitions on harmful AI applications.

Key Developments

The agreement delivers substantial compliance timeline extensions for high-risk AI systems (HRAIS). For Annex III systems (use-based, like those in employment screening or loan decisions), obligations have been postponed from August 2, 2026 to December 2, 2027—a 16-month reprieve. Annex I systems (product-regulated, including medical devices) gain a one-year extension, moving from August 2, 2027 to August 2, 2028.

Simultaneously, the agreement introduces tough new prohibitions effective December 2, 2026. “Nudifier” applications—AI systems that generate, manipulate, or deepfake sexually explicit or intimate images, video, or audio without explicit consent—are now explicitly banned. The prohibition also covers systems creating child sexual abuse material, reflecting growing international concern over synthetic intimate content abuse.

The reformed AI Office gains exclusive supervision authority for AI systems based on general-purpose AI (GPAI) models when the same provider develops both the model and the system, with carve-outs for regulated medical devices, critical infrastructure, and law enforcement tools.

Why This Matters for Ireland and Europe

For Irish and European AI developers and deployers, these changes reshape compliance planning. The extended deadlines acknowledge real-world implementation challenges—smaller firms and resource-constrained organisations now have breathing room to audit systems, implement governance structures, and train teams. However, the new nudifier ban signals Brussels’s determination to regulate AI harms swiftly, even as broader frameworks stabilise.

The AI literacy adjustment—shifting from providers “ensuring” staff competency to “taking measures to support” AI literacy development—offers welcome flexibility while maintaining accountability.

Practical Implications

Organisations deploying high-risk systems should immediately review their Annex III and Annex I categorisations to recalibrate timelines. Any company operating or developing generative AI faces urgent decisions: if your GPAI model powers deployed systems, understand the AI Office’s new enforcement scope.

For creators and platforms, the nudifier ban is straightforward: non-consensual intimate image generation tools must be pulled or prohibited outright. Compliance here is binary and rapidly enforced.

Open Questions

How will the AI Office operationally manage supervision of thousands of GPAI-based systems? What constitutes “nudifier” intent—will satire or artistic use receive exception? And crucially: will these amendments become formal EU law by their July publication target, or face parliamentary delays?

Formal adoption is expected in June 2026, with publication in the Official Journal anticipated in July.

Source: Council of the European Union, European Parliament, European Commission