Claude Sonnet 5.5 carries the text watermark, Anthropic's table shows

Published 30 September 2026

Anthropic's help article now lists Claude Sonnet 5.5 as carrying all three of its marks: the text watermark on Anthropic's own surfaces, the text watermark through AWS, Google Cloud and Microsoft Foundry, and C2PA Content Credentials on files. That makes five models with the text watermark: Claude Fable 5.1, Mythos 5.1, Opus 5.5, Opus 5 and Sonnet 5.5[1].

What changed

The table headed "Which Claude models support watermarking" gained one row on 28 September 2026. An archived copy of the article from 10:26 UTC that day does not have it, and the article's own modification time is 18:34 UTC the same day. Nothing else in the article body changed. Sonnet 5.5 carries no footnote, where Opus 5 still has one about its cloud-partner rollout[1]. The table now lists fifteen models for Content Credentials, and ten of them remain listed for files only, among them Sonnet 5, 4.6 and 4.5[1]. Anthropic's developer documentation also added Sonnet 5.5 to the models that support the code execution tool, which is where the API signs generated files[13].

What it means

Text from Sonnet 5.5 is listed as marked both on Anthropic's own surfaces and through the cloud partners, and marking applies wherever Claude is offered, worldwide[1]. Text from Sonnet 5, 4.6 and 4.5 is not listed as marked today. Anthropic says it is adding watermarks to models released before 2 August 2026, with all of them covered by 2 December 2026[1]. So an unmarked Sonnet response still says little on its own. What Anthropic marks, and how.

What is still not known

The table gives no date from which Sonnet 5.5 output carries the mark, and Anthropic has not published detection accuracy for any model. The detector is still in private preview for organisations obliged to check under EU law[1]. What to do if a detector flags your text.

Last verified against primary sources: 2026-09-30