- EU rules from August 2 require AI-generated images, audio, and text to be labeled to combat deceptive deepfakes and protect democracy.
- The rules apply to new AI systems entering the EU market immediately, with existing systems given four extra months to comply.
- Synthetic content designed to look truthful must be visibly marked as AI-generated and include a digital watermark; texts on public interest without human editorial oversight also require labeling.
- Personal content and evidently artistic, satirical, or fictional works are exempt from the rules.
- Non-compliance can result in fines up to €15 million or 3% of global turnover.
- Examples of deepfakes include manipulated audio of a Slovakian politician, fake videos of Emmanuel Macron and Keir Starmer.
- Tech industry groups worry about overly broad definitions leading to label fatigue, similar to cookie banners, burdening commercial activity.
- Major tech companies like TikTok, Google, and Meta already have AI labeling policies; Google's SynthID watermarks over 100 billion images.
- The European Commission's voluntary code of practice, signed by over 180 organizations, helps companies comply; non-signatories must show compliance by other means.
- The rules are part of the broader EU AI Act enforcement, aiming to balance innovation with public protection.