Model Releases
How Claude Watermarks AI-Generated Text
Anthropic’s Claude models embed a subtle statistical watermark in generated text by biasing token choices during decoding. Sebastian Raschka’s 48‑minute lecture walks through this sampling scheme, exp
Anthropic’s Claude models embed a subtle statistical watermark in generated text by biasing token choices during decoding. Sebastian Raschka’s 48‑minute lecture walks through this sampling scheme, explains how the watermark can be detected and removed, and discusses conditions under which it may fail or break. Slides and a cleaned transcript provide detailed figures for deeper technical review.
Related
- Anthropic to start watermarking Claude-generated text, images
- 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)
- 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)
- Claude will apply invisible watermarks to AI text and images
- Anthropic explains how Claude’s invisible text watermarks will work
Source: Sebastian Raschka | 2026-08-22