Introduction of AI Watermark Removers
A market for removing AI watermarks has sprung up after Anthropic disclosed its use of invisible marks in everything Claude writes. This has led to the creation of various tools and services claiming to remove these watermarks, including a GitHub project with over 4,500 stars and several web tools.
Tools and Services
One of the largest tools is watermarks-remover, an MIT-licensed tool from software developer Guillaume Meyer, founder of Memo. It initially began as a Claude-only agent skill but now claims to support removal of watermarks from Claude, Gemini, and SynthID-Text, as well as OpenAI provenance surfaces and open-weight models using Kirchenbauer-style marks.
Other tools and services include claude-watermark-cleaner, remove-ai-watermarks, and noai-watermark, as well as web tools like claudewatermark.com, claudewatermark.rip, gptcleanup.com, and claudewatermarkremover.app. StealthGPT, which sells AI detection evasion, has also added a Claude watermark remover to its use-case pages.
Effectiveness of Watermark Removers
The effectiveness of these tools is uncertain, as Anthropic has not yet published how its watermark works or released a detector to verify the removal of the watermark. While some tools claim to remove hidden characters and metadata from files, this is not a significant achievement, as file metadata does not survive a re-save, format conversion, or screenshot.
The hard part is removing the actual watermark, which lives in the words chosen by the model. The only known way to remove it is to rewrite the text heavily using a second model. Meyer's tool, for example, only removes metadata for now and may add the ability to strip the actual marks later.
Commercial Sites and Independent Testing
Commercial sites are less careful, with some promising clean, undetectable output, but their claims are measured against ordinary AI detectors rather than Anthropic's watermark. Independent testing has already found gaps in some of these tools, with one popular text cleaner letting the most common hidden-payload technique through untouched.
Why Claude is Marking Text
Anthropic has published a support page explaining its approach to marking text. The company is using an imperceptible watermark woven into the wording of text generated by its models, including Claude, Gemini, and SynthID-Text. This is in response to the EU AI Act, which requires transparency in AI-generated content.
The watermark is applied at the model level and appears across the API, claude.ai, Claude Code, Claude Cowork, and Claude Tag, as well as through AWS, Google Cloud, and Microsoft Foundry. The trigger for the watermark is Article 50 of the EU AI Act, which is enforceable since August 2, 2026, with penalties reaching 15 million euros or 3% of global turnover.
Reaction and Concerns
The reaction online has been mixed, with some users calling watermarking a "compliance checkbox" rather than a real defense. Others have expressed concerns about the effectiveness of the watermark and the potential for it to be removed or evaded.
Defenders should be worried about the potential for a fast-moving category of tools that people wire directly into agent pipelines and then feed their documents through, creating a supply chain surface. The next wave of tools may not be as open and readable as the current ones, and buyers may not be able to verify any claims.
Conclusion
In conclusion, the emergence of AI watermark removal tools is a significant development in the field of AI-generated content. While some tools claim to remove watermarks, their effectiveness is uncertain, and the potential for evasion or removal is a concern. As the market for these tools continues to grow, it is essential to monitor their development and ensure that they do not compromise the integrity of AI-generated content.
Source: BleepingComputer