Anthropic details Claude's text watermark: it only shows Claude was likely involved, is sparse in code and factual text, and disappears after a full rewrite (Anthropic)

Anthropic has detailed Claude's text watermark, which indicates the likelihood of Claude's involvement in generating text. The watermark is sparse in code and factual text, and it disappears completely after a full rewrite. This feature is designed to show that Claude was likely involved in the text's creation, rather than definitively proving it, and it will be present in future Claude models.