How to Remove the Watermark from Claude Sonnet 5.5 Output

To remove the watermark from Claude Sonnet 5.5 output, rewrite the text through a different, non-watermarking model so the word-level pattern no longer matches. Sonnet 5.5 does watermark its text: Anthropic lists it in the model table on its support page, even though the launch announcement never mentioned it.

You can do the rewrite in one pass with ClaudeRewrite. First, here is what is known, so you are not cleaning text on a false assumption.

Is Sonnet 5.5 watermarked? Yes

Claude Sonnet 5.5 launched on September 28, 2026, after the August 2, 2026 cutoff. Anthropic's Sonnet 5.5 launch page does not mention watermarking, but the model table on Anthropic's support page lists a text watermark for Sonnet 5.5 in Claude, Claude Code, Claude Cowork and the Claude API, and through AWS, Google Cloud and Microsoft Foundry. Files it creates also carry Content Credentials. Our Sonnet 5.5 page has the details.

Updated November 12, 2026: an earlier version of this post said the status was not confirmed, because we had only the launch page to go on.

How to clean Sonnet 5.5 output

1. Paste the text into ClaudeRewrite. It first strips invisible characters (zero-width spaces and similar leftovers from chat interfaces). These are not the watermark, but they are a common AI tell.
2. Let it rewrite the wording. The text is rewritten with DeepSeek V4 Flash, an open-weight model that carries no watermark of its own. This is the step that changes the word-level pattern, because the statistical watermark lives in word choices, not in hidden characters.
3. Check your facts. Names, figures and quotes should survive a rewrite, but read the result before you publish. A rewrite is a second pass, not a fact-check.
4. Strip metadata on files. If you are sharing a document or image, ClaudeRewrite also removes C2PA, EXIF and XMP metadata.

What Anthropic says about editing

Anthropic says the watermark travels with text that is copied and pasted and may persist through some editing. It also lists text that has been heavily edited, paraphrased or translated, and very short passages, among the cases where no mark may be detected. That is why light edits are not a dependable fix, and a full rewrite changes far more of the wording.

What this cannot promise

Anthropic's watermark detection is in private preview for eligible organizations and for enterprises checking their own Claude output, so there is no public way to check Sonnet 5.5 text before or after cleaning. We cannot show you a before-and-after detector score, and we do not claim one. Treat a rewrite as reducing risk, not as a guarantee.

Related

• Is Claude Sonnet 5.5 watermarked?
• Does Claude Opus 5.5 watermark its text?
• How to remove Claude's watermark (full guide)

References

• Anthropic support: How Claude marks AI-generated content
• Anthropic: Introducing Claude Sonnet 5.5
• Anthropic: How Claude's text watermarking works