Claude’s invisible watermark could expose AI-written work

Anthropic announced it is adding invisible watermarks to text and images generated by its AI chatbot Claude to help identify AI-created content, aligning with the European Union's AI Act. The text watermarking uses subtle statistical word choices detectable by automated systems without altering readability, but Anthropic acknowledged that a detected mark confirms processing by Claude but does not conclusively prove full provenance, raising concerns about false flags on human-edited work.

Claude’s invisible watermark could expose AI-written work Anthropic /tech/claude-anthropic-auto-mode-update-new-b3030616.html announced it is adding invisible watermarks to text and images generated by its AI chatbot Claude to help identify AI-created content.- The global update aligns with requirements under the European Union's AI Act, which mandates methods for detecting AI-generated material. - Text watermarking relies on subtle statistical word choices detectable by automated systems without altering the readability or meaning of responses. - The feature raised concerns among users that Claude-edited human-written work might be falsely flagged as AI-generated content. - Anthropic acknowledged system limitations, stating that a detected mark confirms content was processed by Claude but does not conclusively prove full provenance.