Inspect content you own for listed invisible Unicode characters and supported file metadata. Each local check shows what it found—and what that result cannot prove.
Independent tool. Not affiliated with, endorsed by, or sponsored by Anthropic.
Three separate mechanisms
“Claude watermark” can refer to different things. This tool treats them separately: hidden Unicode characters, statistical patterns in text, and file metadata or provenance. Finding or removing one does not prove anything about the others.
Choose a mode
Mode A · local text check
Paste text or choose a supported text file. Review every finding before removing anything.
Scans for the listed hidden or unusual Unicode characters in this browser. Removing these characters does not detect or remove a statistical text watermark.
Before you begin: Only process content you own or have permission to use. Do not use this tool to misrepresent authorship or evade rules that apply to you.
Your text is processed in this browser and is not sent to us for this mode.
Mode B · local file check
Choose a file you own or are authorized to inspect. Results remain scoped to supported fields.
Inspects supported metadata fields in this browser. A “not found” result is not proof that content is human-created.
JPEG, PNG, PDF, or DOCX; one file; maximum 10 MiB.
Your file is processed in this browser and is not uploaded to us for this mode. Its name is not displayed or logged.
No file selected.
Mode C · fail closed
Rewrite is not available. A rewrite may change statistical patterns, but no tool can promise a particular detector result. This feature will remain off until its provider and privacy terms are verified.
This tool cannot guarantee detection or removal of any statistical watermark, cannot verify results against Anthropic's private detector, and cannot guarantee that content will be classified as human-written or accepted by any detector, school, employer, publisher, platform, or reviewer.
Result guide
Invisible code points can affect formatting, search, copy-and-paste behavior or text review. The scanner identifies the listed characters by position, Unicode code point, category and name. You choose what to remove.
This does not detect or remove a statistical text watermark.Files may contain EXIF, XMP, document properties, C2PA or Content Credentials data. The inspector reports fields supported by the implemented parser and format.
A field reported as not found is not proof of human authorship or the absence of other provenance information.A statistical watermark can be created through token-selection patterns during generation. It is not a hidden character and it is not file metadata. This mode remains unavailable.
No detection, removal or third-party acceptance outcome is promised.Process
Use the text scanner for listed Unicode characters or the file inspector for supported metadata fields. The checks are not interchangeable.
Text findings show code points, categories and positions. File findings show supported field families. The page does not combine them into a detector score or authorship result.
Remove reviewed Unicode characters from the working copy, or use the metadata inventory to guide your own file review. Keep an original copy and check the result before publishing or sharing it.
Plain-language boundaries
This utility is for reviewing content you own or are authorized to inspect. It does not certify authorship, originality, policy compliance, or acceptance by a third party.
Third-party option
A separate third-party project, guillaumemeyer/watermarks-remover, describes agent skills and a Python service for Unicode inspection, file-provenance workflows, and optional rewrite workflows. Its documentation also describes a Claude Code plugin and a self-contained text-only skill. It is independent software, is not part of this browser tool, and does not give this page access to an Anthropic detector. Source documentation reviewed September 16, 2026 at commit 21838ae922bb07a0c734e57df53fd40007561bb3 under the MIT License. Review that exact source and the project's current code, permissions, setup, and documentation before installing or running it.
Questions
Coverage changes. Check Anthropic’s current documentation for the model and output type you use.
Anthropic’s coverage table separates text watermarks from Content Credentials (C2PA) and varies by model and delivery surface. As of September 16, 2026, some listed Claude models support text watermarks while all listed models support C2PA in generated files; coverage can change, so check the source for the current table. Source: Anthropic, “How Claude marks AI-generated content” (updated September 16, 2026; reviewed September 17, 2026).
It can remove the listed Unicode characters. That is separate from statistical text watermarking and file provenance metadata.
Detector access and availability can change. Check Anthropic’s current documentation. This independent page does not have access to an Anthropic detector and is not an official or equivalent detector.
A substantial rewrite may disrupt statistical patterns, but this tool cannot verify that against Anthropic's private detector and does not guarantee removal. Rewriting can also change meaning, facts, citations or tone, so every change needs review.
At the reviewed September 16, 2026 commit, the independent guillaumemeyer/watermarks-remover project describes agent skills and a Python service for Unicode inspection, file-provenance workflows, and optional rewrite workflows. It is not part of this site and does not give this page access to an Anthropic detector. Review the exact linked source and the project’s current code, permissions, setup, and documentation before installing or running it.
Do not assume every Claude-generated file contains C2PA or Content Credentials data. This inspector reports supported fields in supported file types. “Not found” means only that this version did not find the field it checked; it does not prove human authorship or rule out other provenance information.
No. Hidden Unicode, statistical patterns and file metadata are different mechanisms, and tools have limited visibility. This tool cannot guarantee detection or removal of any statistical watermark, cannot verify results against Anthropic's private detector, and cannot guarantee that content will be classified as human-written or accepted by any detector, school, employer, publisher, platform, or reviewer.
Privacy and control
Unicode scanning and supported file inspection run in this browser. We do not receive or store your text or file contents for these local modes.
Refresh or close the page to clear the tool’s in-page working state. Copies you create, downloads, browser/OS caches, and clipboard contents remain under your control.
Read the site Privacy Policy, Terms, third-party notices, or contact us with questions.
Local checks only
Choose a local text scan or local file inspection. Review each finding in context before changing your content.
By using the tool, you confirm that you own the content or have permission to process it and will follow rules that apply to you.