Anthropic details Claude’s invisible text watermarking for EU AI Act
TL;DR. Anthropic outlined how Claude will embed invisible watermarks into its generated text to meet new European Union AI transparency regulations. - Claude’s text marking system is a variant of Google DeepMind’s open-source SynthID-Text technology. - This system uses wording probability patterns to create detectable, machine-readable content indicators. - The watermarking helps comply with EU AI Act requirements for synthetic audio, image, video, and text transparency.
- Anthropic is implementing invisible text watermarks for Claude-generated content.
- The system is a modified version of Google DeepMind's open-source SynthID-Text.
- Watermarks utilize wording probability patterns to make AI-generated text detectable.
- This initiative aims to comply with the European Union's AI Act transparency mandates for synthetic media.
Sources
- Anthropic explains how Claude’s invisible text watermarks will work — theverge.com
- the-decoder.com — the-decoder.com
- medium.com — medium.com
- mashable.com — mashable.com
- theguardian.com — theguardian.com
- daringfireball.net — daringfireball.net