Claude text watermark

Summary

Since 2026-11-03 (platform release notes), text generated by Claude Mythos 5, Claude Opus 4.7, and Claude Sonnet 4.6 carries Anthropic's text watermark on the platforms listed in Anthropic's support article. The docs say it does not change meaning, quality, or readability, adds no tokens or hidden characters, and carries no information about the user or organisation.

Key points

• No action needed: requests and response handling stay the same.
• Sampling caveat: on Sonnet 4.6, output for a request that sets temperature, top_p, or top_k to a non-default value might not carry the watermark. If you rely on it, omit all three, including any a framework sets for you.
• Models covered: the release note names Mythos 5, Opus 4.7, and Sonnet 4.6, not the 5.5-generation models; the support article holds the full, current list.
• Detection: the docs page does not describe a public detector or API; background is in Anthropic's "How Claude's text watermark works" post, not read for this note.

Why it matters

A statistical watermark lets text be identified as Claude-generated without visible markers. Teams that need provenance should avoid custom sampling settings on Sonnet 4.6.