Overview
- EU negotiators, who closed a provisional deal early Thursday, pushed many high‑risk AI obligations to December 2, 2027.
- High‑risk covers systems used for biometrics and for decisions in critical infrastructure, schools, hiring, policing, and border checks, which the EU says need more time and standards before enforcement.
- Industrial machinery using AI will follow sector safety law instead of the AI Act after a late German push, while other sectors such as medical devices remain under the AI rules.
- The deal creates an explicit EU ban on AI systems that generate non‑consensual sexual images or child sexual abuse material, covering images, video, and audio, with companies required to comply by December 2, 2026.
- Watermarking of AI‑generated content becomes mandatory on December 2, 2026, and the agreement still needs formal approval by the European Parliament and the Council before the summer.