EU Parliament and Member States Reach Provisional Deal to Amend AI Act
The tentative agreement, part of the Commission's digital omnibus package, introduces strict prohibitions on unauthorised sexually explicit deepfakes while postponing compliance dates for biometric and critical infrastructure systems to December 2027.
European Union member nations and Parliament lawmakers have reached a provisional agreement to amend the EU AI Act, a move designed to simplify the landmark legislation while addressing emerging security risks. The deal, reached following late-night negotiations, forms part of the European Commission's broader digital omnibus package aimed at streamlining the regulatory framework. While the agreement is currently tentative, it requires formal approval from the European Parliament's plenary session and EU governments to become law.
A central component of the amended text is a specific prohibition on AI applications used to create unauthorised sexually explicit deepfakes and imagery. This ban covers images, video, and audio, and explicitly includes material depicting child sexual abuse. Companies developing these systems have until December 2 this year to bring their operations into compliance with the new restrictions. The decision is viewed as a direct response to recent incidents where users utilised Elon Musk's chatbot Grok to generate and disseminate millions of such images online.
In a significant shift regarding implementation timelines, the deadline for high-risk AI systems has been postponed. Obligations concerning systems involving biometrics, critical infrastructure, education, employment, law enforcement, and border management are now set to commence on December 2, 2027, rather than the original August 2 date. Critics have interpreted this delay as a concession to business pressure intended to ensure legal certainty and reduce administrative costs for the sector.
The agreement also sees the exclusion of machinery from the scope of the AI Act. This adjustment is based on the rationale that such equipment already complies with existing sectoral safety rules and does not require the additional regulatory burden of the new legislation. Meanwhile, the schedule for mandatory watermarking of AI-generated content remains unchanged, with the requirement set to apply from December 2, 2026.
Marilena Raouna, Deputy Minister for European affairs of Cyprus, which currently holds the rotating presidency of the EU Council, defended the amendments as supportive of the region's digital sovereignty. She stated that the agreement reduces recurring administrative costs and ensures a smoother, more harmonised implementation of rules across the Union. The European Commission has framed these changes as evidence of the institutions' ability to act swiftly to protect children and deliver on commitments.
Despite the progress, the nature of the deal remains provisional until the final legislative hurdles are cleared. The exact scope of the prohibition on unauthorised sexually explicit imagery in the final legal text may require further interpretation. Nevertheless, the current consensus among negotiators suggests that the amendments will proceed to formal approval, marking a significant evolution in how the bloc governs artificial intelligence.
