The European Union has agreed to slow down the most demanding parts of its landmark AI Act and add a fresh prohibition on AI tools that create non-consensual intimate imagery. The provisional deal, reached on 7 May between the Council presidency and European Parliament negotiators, forms part of the Commission's "Digital Omnibus" simplification package and now goes to formal adoption.
Under the agreement, the application date for stand-alone high-risk AI systems (the Annex III category) moves from 2 August 2026 to 2 December 2027. High-risk AI embedded in regulated products (Annex I) gets a longer runway, with rules now applying from 2 August 2028. National authorities have also been given until 2 August 2027 to set up the regulatory sandboxes the original text required earlier.
A new prohibited practice
The headline addition is a new prohibited practice. From 2 December 2026, AI systems used to generate non-consensual sexual or intimate content, including child sexual abuse material, will be banned outright. The provision is aimed squarely at "nudifier" apps and adjacent forms of intimate deepfake abuse, an area that has drawn growing political pressure across member states.
Other changes are more technical. The grace period for providers to implement transparency markings on AI-generated content has been cut from six months to three, with the new deadline set at 2 December 2026. The AI Office's powers over general-purpose AI models have been clarified, with national authorities retaining competence in defined areas such as law enforcement, border management, judicial matters and financial supervision. Exemptions previously granted to SMEs have been extended to small mid-caps, and developers have been given more room to process sensitive personal data specifically for bias detection and mitigation.
Relief or retreat
Reactions have split along familiar lines. The Commission has framed the deal as targeted relief, designed to give companies time to comply without weakening the law's substance. As Euronews reported on 21 May, supporters in the Parliament argued that delaying the high-risk regime gives standardisation bodies the time they need to publish technical specifications that are currently incomplete. Critics counter that postponement is postponement, and that high-risk uses of AI in employment, education and essential services will now go unregulated for an extra year and a half.
The agreement still needs formal adoption by both the Council and the Parliament before it enters force. Once it does, the EU will have a slightly slower AI Act, a slightly broader list of banned practices, and a one-year head start on the United States, where the Trump administration recently shelved its own draft AI safety executive order after industry lobbying.
Commentarii · 0