3 outlets·3 reports
Anthropic explains how Claude’s invisible text watermarks will work
ThinkingNews Desk · how this was written
Anthropic has implemented an invisible text watermark in Claude that works by subtly adjusting word-choice probabilities to embed a unique fingerprint in generated content. The company says the watermark does not affect the model’s output quality, though analysts note the alteration could degrade the writing.
Written from all 3 reports below, not from any single one.
How it was reported
- Hacker News·Anthropic's 'Watermark' Text Adulteration in Claude Is a Perversion of Writing
- TechMeme·Anthropic's text watermark alters word probabilities to embed a fingerprint, which could degrade Claude's writing, despite its claim of no impact on quality (John Gruber/Daring Fireball)
- The Verge·Anthropic explains how Claude’s invisible text watermarks will work