Overview
- The European Commission’s final guidelines set transparency rules that take effect on Sunday, August 2, 2026, requiring visible labels and embedded machine-readable markers on AI-generated or AI-manipulated audio, images, video and text.
- The guidance creates a two-tier disclosure approach that lets clearly artistic, fictional, or satirical works use less intrusive labeling while requiring clear, distinguishable notices at first exposure for content that could be mistaken for reality.
- The obligations apply broadly and can reach non-EU providers or deployers when their AI outputs are used or distributed in the EU, and enforcement can include fines up to €15 million or 3% of global turnover for serious non-compliance.
- Major platforms including Google and Meta have signed the voluntary Code of Practice and are pushing interoperable watermarking and provenance tools such as SynthID and C2PA as practical ways to meet the marking and detectability demands.
- Sectors such as videogames face grey areas where fantasy content is exempt but photorealistic recreations of real people or events must be judged case by case, prompting firms to map AI use cases, train staff, and adopt metadata and watermark standards to avoid penalties.