Per-model status
Is Claude Haiku 5.5 watermarked?
Yes. Claude Haiku 5.5 launched on October 7, 2026, after Anthropic's August 2, 2026 cutoff, and Anthropic's support page lists it as carrying the text watermark in Claude, Claude Code, Claude Cowork and the Claude API and through AWS, Google Cloud and Microsoft Foundry.
Haiku 5.5 is Anthropic's small model for high-volume tasks such as summaries and classification. Anthropic notes that a very short passage can leave too little text for a reliable signal, which limits detection of short outputs. The launch announcement does not mention watermarking; the model table on Anthropic's support page does. Files Haiku 5.5 creates also carry Content Credentials (C2PA).
Updated November 12, 2026. This page previously listed Haiku 5.5 as not confirmed, because its launch announcement does not mention watermarking.
Where Anthropic marks Haiku 5.5 output
- Text watermark in Claude, Claude Code, Claude Cowork and the Claude APIYes
- Text watermark through AWS, Google Cloud and Microsoft FoundryYes
- Content Credentials (C2PA) on files it createsYes
From the model table on Anthropic's support page, checked November 12, 2026.
FAQ
Is Claude Haiku 5.5 watermarked?
Yes. Anthropic's support page lists Claude Haiku 5.5 as carrying its text watermark in Claude, Claude Code, Claude Cowork and the Claude API and through AWS, Google Cloud and Microsoft Foundry. It launched after the August 2, 2026 cutoff, and Anthropic says models launched on or after that date are marked from launch.
Can anyone detect Claude Haiku 5.5 output?
Not publicly. Anthropic's watermark detection is in private preview. It is available to eligible organizations such as regulators, law enforcement, media, fact-checkers, independent researchers, educational organizations and EU civil society groups, and to enterprises that need to check their own Claude output. Anthropic says it plans to expand access over time. Its free Content Checker checks files for Content Credentials, not text.
Does the Haiku 5.5 watermark survive editing?
Partly, according to Anthropic. It says the watermark travels with text that is copied and pasted and may persist through some editing, but that text which has been heavily edited, paraphrased or translated, or a very short passage, may carry no detectable mark. There is no public detector, so you cannot test a piece of text yourself.
How do I remove the watermark from Claude Haiku 5.5 output?
A complete rewrite through a different, non-watermarking model changes the word choices the watermark lives in. ClaudeRewrite does this automatically using DeepSeek V4 Flash and is designed to keep your facts, numbers and meaning intact. Because there is no public detector, the result cannot be independently verified.