AI Watermark Removers Emerge After Anthropic’s Claude Update
TL;DR. A market for AI text watermark removers appeared after Anthropic announced invisible marks in Claude's output. - Various tools claim to defeat text watermarks, including open-source projects and paid services. - Anthropic has not released a detector, making verification of these tools' effectiveness impossible. - Watermark removal primarily involves metadata stripping or extensive text rewriting, not inherent mark deletion.
- Anthropic implemented invisible watermarks in all text generated by its Claude models.
- A market for 'watermark removers' has quickly emerged, including open-source tools and commercial services.
- The effectiveness of these removal tools against actual AI watermarks cannot be verified without Anthropic's detector.
- True watermark removal requires heavy text rewriting, as the mark is embedded in word choice, not metadata.
Sources
- AI 'watermark removers' flood the web. Almost none can prove they work. — bleepingcomputer.com