Anthropic Embeds Invisible Watermarks in Claude's Text to Comply with EU AI Act
Summary
Anthropic is embedding invisible watermarks into Claude's AI-generated text using Google DeepMind's SynthID-Text system to comply with the EU AI Act's mandate that synthetic content be machine-readable identifiable, with no impact on output quality or user costs.
Key Points
- Anthropic is introducing invisible text watermarks to Claude using a version of Google DeepMind's open-source SynthID-Text system, which embeds detectable patterns into AI-generated text through word probability choices.
- The watermarking is being implemented to comply with the European Union's AI Act, which mandates that synthetic text, audio, images, and video contain machine-readable marks identifying them as artificially generated.
- Anthropic confirms the watermarking will not increase costs for users or affect the quality of Claude's outputs, and other major AI developers including Google and OpenAI are also subject to the same EU transparency requirements.