NewsBite analysis · AI, Security · Open Source
ProPakistani
New Free Tool Removes Claude Watermark a Day After Anthropic Announcement
In response to Anthropic's recent announcement regarding its invisible watermarking system for Claude-generated text, developers created an open-source project called watermarks-remover. This tool targets various AI markers, including those from Claude, Google’s Gemini and SynthID, and the provenance formats from OpenAI. While the tool offers methods for removing or weakening these markers, especially the statistical watermark, it operates on a best-effort basis without guarantees of complete success. Anthropic specifies that its watermark does not use visible Unicode characters and is crafted to subtly alter word selection in generated text. Although the watermarks-remover tool provides layers for removing certain markers, the statistical watermark is complex, requiring substantial text rewriting for effectiveness. This kind of alteration may impact the quality of the output, raising questions about whether heavily modified text retains its original AI-generation attribution.
ExperimentConfidence: MediumShort term horizon
Published 19 Aug 2026, 10:55 UTC2 min read
Visit original article ↗Lead image — ProPakistani
Summary
TL;DR
- New tool removes or weakens AI provenance markers.
- Developed shortly after Anthropic's watermarking system announcement.
- Targets multiple AI systems including Claude and OpenAI.
- First layer removes direct markers while the second weakens statistical watermarks.
- Anthropic cautions that minor edits won't eliminate the watermark.
Interesting facts
- The tool can strip metadata from various file formats including PDF and DOCX.
- Anthropic is preparing a watermark detection API for users to verify text provenance.
- The effectiveness of watermark removal varies and can't guarantee success.