How Claude’s invisible watermarks work
Anthropic adds invisible text watermarks to Claude using Google's SynthID-Text approach.
The update
Anthropic is adding invisible text watermarks to Claude to meet the EU AI Act’s transparency requirements. The system uses a version of Google DeepMind’s open-source SynthID-Text approach, which embeds detectable patterns into word choices. The watermarking does not affect output quality or cost, and watermarked text is indistinguishable to readers. Anthropic plans to release a watermark detection API to enable verification of AI-generated content.
Why it matters
As AI-generated content proliferates, tools that can reliably identify it become critical for academic integrity, journalism, and content verification. This move positions Anthropic as a leader in transparent AI development and sets a precedent for other major AI providers who must also comply with the EU’s transparency requirements.
What to watch
Users should test whether light editing removes the watermark, as Anthropic notes that while “light editing probably won’t remove the watermark completely,” a complete rewrite where every word is replaced will. The system’s effectiveness with code generation and how it compares to other AI watermarking efforts remain open questions.
Sources
- theverge.com — overview of SynthID-Text and EU AI Act compliance
- techcrunch.com — details on API plans and editing resistance
