Anthropic's Claude Just Got a Hidden Watermark That Follows Your AI-Written Text Everywhere You Paste It

Anthropic has introduced imperceptible, machine-readable watermarks into text generated by its Claude models launched on or after August 2, 2023. These watermarks, which do not alter the text's meaning, quality, or readability, persist when users copy and paste content and may survive some editing. The feature applies globally across supported models accessed via Claude, Claude Code, Claude Cowork, Claude Tag, Anthropic’s API, and third-party platforms like AWS, Google Cloud, and Microsoft Foundry. The move aligns with the EU AI Act’s transparency requirements, which mandate identifiable synthetic content. In education, the watermark could help schools detect AI-generated submissions, while publishers may use it to address AI authorship concerns. However, the watermark is not foolproof: heavy editing, paraphrasing, translation, or mixing with human-written text can remove the detectable signal. Additionally, a detected mark does not confirm original AI authorship, as using Claude for proofreading or summarizing human text can also leave a watermark. Anthropic is also working to retrofit older models and develop third-party detection tools. The company recently settled a $1.5 billion copyright dispute with authors, underscoring ongoing tensions over AI-generated content. Google DeepMind’s SynthID watermark for Gemini models is a similar initiative, adjusting token probabilities without affecting output quality.

(Source：Inkl)