The European Commission announced on 7 May 2026 that EU member states have agreed to amendments simplifying portions of the AI Act while expanding prohibited practices to include so-called "nudification" applications [source]. The changes represent the first substantive revision to the regulation since it entered into force in 2024.

Under the agreed text, AI systems that generate or manipulate image, audio, or video content to create sexualised depictions of identifiable individuals without consent will be classified as prohibited practices under Article 5 of the AI Act. Providers deploying such systems within the EU will face administrative fines of up to €35 million or 7% of global annual turnover, whichever is higher.

The amendments also introduce streamlined compliance pathways for general-purpose AI models below specified capability thresholds. Models that do not meet the systemic risk criteria defined in Article 51 will be exempt from certain transparency obligations previously required under Annex XII, including detailed technical documentation on training data provenance and model evaluation protocols.

The European AI Office stated that the changes aim to reduce administrative burden on smaller developers while addressing emerging harms observed since the Act's initial implementation. The office cited a rise in reports of non-consensual intimate imagery generated by widely available image synthesis models as a driver for the nudification ban.

Member states have six months to transpose the amendments into national law. The European AI Office will publish updated guidance on prohibited practices and compliance requirements by the end of Q2 2026. Providers currently offering systems that fall under the new prohibitions must cease EU operations or modify their services to comply before the transition period expires in November 2026.