Bottom line: Shortly after Anthropic introduced a text watermark for Claude, several circumvention tools have emerged whose effectiveness cannot be verified due to the lack of a public detector.
Just days after Anthropic began watermarking text generated by Claude, numerous tools claiming to be able to remove this watermark are already circulating. None of these claims can be substantiated, since Anthropic has not yet provided a public detector.
Shortly after Anthropic introduced the text watermark for Claude outputs, several so-called “watermark removers” have appeared. Among them is an open-source project on GitHub with over 4,500 stars, as well as several paid services positioning themselves as tools for AI Detection Evasion. The offerings range from freely available code to commercial services.
The problem is that none of these tools’ effectiveness claims can be verified. Anthropic has not yet published its own detector that could prove whether a text actually carries a Claude watermark or whether it has been removed by one of the circulating tools. Without such a reference point, there is no reliable basis for the providers’ marketing claims.
For security officers in enterprises, this creates a twofold uncertainty: On the one hand, the origin of texts that may have been created with Claude cannot currently be reliably verified via watermarks. On the other hand, an ecosystem of tools is emerging whose functional promises are unvalidated, which calls for caution both in internal compliance processes and when assessing AI-generated content in supply chains or communications. Anyone planning to use watermarks as a control mechanism for proving AI-generated content should consider that this protective measure has already become the target of active circumvention attempts shortly after its introduction — regardless of whether these attempts actually work.
Until Anthropic provides a verifiable detector, the watermark’s evidentiary value for forensic or compliance purposes remains limited. Organizations that rely on provenance verification for AI-generated texts should factor this into the risk assessment of relevant use cases.
Source: www.bleepingcomputer.com · Published August 13, 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrasing and classification by Lumi News Pipeline v1.8.3.