How Claude marks AI-generated content (Help Center)

Summary

Anthropic's Help Center article lists which Claude models carry the text watermark on which platforms, and how Content Credentials (C2PA) and detection work. It shows "Updated today" with no explicit date, so date_published is the date it was read. It answers the question left open by the 2026-11-03 release note, which named only three models.

Key points

First-party surfaces (Claude Platform API, Claude apps, Claude Code, Claude Cowork, Claude Tag): watermarked models are Fable 5.2, 5.1, 5; Mythos 5.2, 5.1, 5; Opus 5.5, 5, 4.8, 4.7, 4.6, 4.5; Sonnet 5.5, 5, 4.6, 4.5; Haiku 5.5 and 4.5.
Cloud partners (AWS, Google Cloud, Microsoft Foundry): Fable 5.2, 5.1, 5; Mythos 5.2, 5.1, 5; Opus 5.5, 5, 4.8, 4.7; Sonnet 5.5, 5, 4.6; Haiku 5.5. Opus 4.6, Opus 4.5, Sonnet 4.5, and Haiku 4.5 are marked only on first-party surfaces. Rollout on Amazon Bedrock for Mythos 5, Opus 4.7, and Sonnet 4.6 completes "by November 10".
Models launched on or after 2026-08-02 carry the mark from launch; older models are in a transition period.
On Sonnet 4.6, a non-default temperature, top_p, or top_k may leave output unmarked.
Detection: text watermark detection is a private preview API for eligible organizations (request form; Anthropic plans to widen access). The free "Claude Content Checker" checks files for Content Credentials only, not text watermarks.
Content Credentials (C2PA) apply only to files Claude creates in the Claude apps and the Claude Platform API, including Claude Platform on AWS and Microsoft Foundry.

Why it matters

Almost all current Claude output is watermarked, not just the three models in the first announcement. There is still no public text detector, so a third party cannot check text without the private API.