Overview
- Most transparency and user‑notification obligations of the EU AI Act come into force on 2 August 2026, requiring providers to disclose when a service or content is produced or driven by AI.
- Mandatory watermarking and labeling for synthetic images, audio and video has been delayed until 2 December 2026 to give industry time to adapt.
- The strict technical and governance rules for high‑risk systems are postponed until December 2027, and systems built into already regulated products have until August 2028 to comply.
- The law already bans a set of 'unacceptable' AI uses such as subliminal manipulation, social scoring and mass facial‑data scraping, and will add non‑consensual sexual deepfakes and AI‑generated child sexual content to the prohibitions in December.
- Spain has moved a domestic implementing bill through government toward parliament and national authorities can now enforce the Act with fines reaching up to €35 million, while experts warn companies might treat the staged rollouts as permission to delay full compliance.