AI 'watermark removers' flood the web. Almost none can prove they work.
Days after Anthropic began watermarking Claude's text, numerous "watermark remover" tools emerged, including a popular open-source project and paid evasion services, but none can prove their effectiveness since Anthropic hasn't released a detector.
Days after Anthropic began watermarking text from its Claude AI, multiple so-called "watermark removers" have appeared online, including an open-source project with over 4,500 GitHub stars and paid detection-evasion services. None of the tools' claims can be verified, as Anthropic has not released a public detector to test whether watermarks are actually removed. The flood of unproven removers highlights the difficulty of enforcing AI-generated text detection.