SafetyPage

Lab system cards and frameworks · Article · Anthropic

Claude text watermark

Read at anthropic.com ↗Read at sourceAll rights reserved

Outline · 54 paragraphs

  1. What is watermarking?¶9
  2. How does watermarking affect Claude’s outputs?¶13
  3. Which specific method of watermarking do you use?¶19
  4. What about cases where Claude has proofread or edited human text?¶23
  5. What about code?¶25
  6. What does this mean for users?¶29
  7. Does this slow the model down, or make it more expensive?¶30
  8. Can a watermark be traced back to me or my organization?¶32
  9. Why are you watermarking Claude’s outputs?¶34
  10. Other questions¶36
  11. How do I check if a piece of text was written by Claude?¶37
  12. What about images and other files?¶39
  13. Can’t someone just edit the text to get around the watermarking?¶42
  14. What does a watermark actually prove?¶44
  15. Do watermarks apply to translations?¶46
  16. What about older Claude models?¶48
  17. How does this differ from AI detection software, like Pangram?¶50
  18. Does this change who owns a given output, or who is legally responsible for it?¶52