Source: Euronews (EU News)

After negotiations that resumed and concluded in the early hours of 7 May 2026 in Brussels, the European Parliament and the Council provisionally agreed an amendment to the EU Artificial Intelligence Act as part of the Digital Omnibus package. The compromise fully incorporates the Parliament’s March proposal to ban so‑called nudifier apps—AI systems that alter images of real people without consent to depict them nude or sexually explicit—and makes both users and companies liable where apps are intentionally designed for this purpose or where companies applied insufficient prevention and security measures. The ban is scheduled to enter into force on 2 December 2026; the text does not list specific apps but the agreement references the January case involving Grok on X as a motivating example, cited by co‑rapporteur Arba Kokalari.

The deal also eases implementation timing and scope for other AI Act obligations: entry into force of new safety standards for AI systems classified as “high‑risk” has been postponed, with listed high‑risk systems now subject to compliance by 2 December 2027 and certain “security components” by 2 August 2028, while the Machinery Regulation will be exempted from the directive. The amendment narrows the definition of “security components,” extends deadlines for watermarking requirements, permits limited uses of personal data to correct biases “only when strictly necessary,” envisages exemptions for small and medium‑sized enterprises, and aims to centralise oversight of general‑purpose AI models under the European AI Office. The agreement is provisional and must be formally adopted by Parliament and Council; the Parliament’s press release sets a completion target of 2 August 2026. Statements from the Cypriot Presidency and the President of the Commission welcomed the compromise as supporting companies and strengthening protections for minors.