# Anthropic Watermarks Claude's Text Output by Biasing the Sampler, Not by Inserting Characters

The method builds on Google DeepMind's SynthID-Text and applies to models launched on or after 2 August, driven by transparency commitments Anthropic signed under European Union rules in July.

- Published: 2026-09-08T06:18:45.692Z
- Canonical: https://polylog.news/ai/2026-09-08/anthropic-watermarks-claude-s-text-output-by-biasing-the-sam
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Anthropic News](https://www.anthropic.com/news/claude-text-watermark), [Anthropic Research](https://www.anthropic.com/research/team/societal-impacts)

Anthropic published an explanation of how Claude's text watermark works. The mark is not a hidden character or an edit added after the fact. It changes the source of randomness the model uses when choosing among candidate next words, produc…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-09-08/anthropic-watermarks-claude-s-text-output-by-biasing-the-sam (subscription information: https://polylog.news/pricing).