What Claude's Text Watermark Actually Does
Anthropic published an explainer on how text watermarking will work in future Claude models. Everything below comes from that announcement on anthropic.com, bylined August 2026.
Transcript
Anthropic says future Claude models will produce watermarked text to comply with a European Union rule.
Claude's watermark is a version of Google DeepMind's SynthID Text approach. It changes only the source of the randomness used to choose between equally good words.
The company says readers cannot distinguish watermarked text from normal output, and that it carries no identifying information about a person or organization. Detection tooling is planned.
The watermark can only say Claude was likely involved, and detection is weaker on short passages and code. Anthropic describes its own system with no outside verification.
Inboxsmith builds AI receptionists that answer calls and messages for small businesses. Please like and subscribe for more news.
Sources
Every claim in this video comes from the top ranking coverage of this topic. The claims and where each one came from:
- Future Claude models will generate text that contains a watermark.(Anthropic official announcement, How Claude's text watermark works)
- This is a way of determining the likelihood that Claude was involved in writing the text, and we, along with several other major AI providers, are implementing this change to comply with the EU AI Act.(Anthropic official announcement, How Claude's text watermark works)
- Claude's text watermark is a version of the SynthID-Text approach published by Google DeepMind.(Anthropic official announcement, How Claude's text watermark works)
- The watermark only changes the source of the randomness used to pick among words.(Anthropic official announcement, How Claude's text watermark works)
- Watermarking does not impact the quality of Claude's output. To a reader, a watermarked response is indistinguishable from an unwatermarked one.(Anthropic official announcement, How Claude's text watermark works)
- Watermarking carries no identifying information and can't be traced to a specific person, organization, or chat.(Anthropic official announcement, How Claude's text watermark works)
- We will soon be offering a watermark detection API. We're in the process of working out the details of its implementation.(Anthropic official announcement, How Claude's text watermark works)
- Detecting a watermarking also doesn't work well on small samples, where there are fewer word choices and thus less information to go on. Code, which in very many cases has to be exact, has generally less watermarking than some other forms of text. A watermark can only determine that Claude was likely involved with the content at some point.(Anthropic official announcement, How Claude's text watermark works)