Built by Pluto.Studio · find and test independent Android apps on IndieHangar
AI Watermarks

Claude Haiku 5.5 carries the text watermark from launch day

Published 8 October 2026

Anthropic's help article now lists Claude Haiku 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. The row appeared on 7 October 2026, the day Anthropic announced the model. That makes nine models with the text watermark: Claude Fable 5.1, Fable 5, Mythos 5.1, Opus 5.5, Opus 5, Opus 4.8, Sonnet 5.5, Sonnet 5 and Haiku 5.5[1].

What changed

The table headed "Which Claude models support watermarking" gained one row. The copy we read on 7 October did not have it; the article's modification time is 17:55 UTC that day. Nothing else in the article body changed. Haiku 5.5 carries no footnote, where Fable 5, Opus 4.8 and Sonnet 5 still have one saying their rollout on Amazon Bedrock will complete by 12 October[1]. The table now lists sixteen models for Content Credentials, and the same seven remain listed for files only: Mythos 5, Opus 4.5, 4.6 and 4.7, Sonnet 4.5 and 4.6, and Haiku 4.5[1]. Anthropic's developer documentation also added Haiku 5.5 to the models that support the code execution tool, which is where the API signs generated files[13].

What it means

Anthropic's rule is that models launched on or after 2 August 2026 support marking at launch, and the table followed the announcement within hours, as it did for Sonnet 5.5 on 28 September[1]. The usual caveats hold: a detected mark means Claude may have processed the text, not that it wrote it, and the detector is still in private preview for organisations obliged to check under EU law[1]. What Anthropic marks, and how.

What is still not known

The announcement itself does not mention watermarking; the help table is the only statement. Anthropic has not published detection accuracy for any model, and still gives no date for finishing the models released before 2 August 2026[1]. What to do if a detector flags your text.

Sources

  1. How Claude marks AI-generated content primaryAnthropic, 2026-08, updated 2026-10-07
  2. Code execution tool — Content Credentials on generated files primaryAnthropic, accessed 2026-09-18

Last verified against primary sources: