Developer Guillaume Meyer published code to remove invisible watermarks from Claude-generated text within four hours of Anthropic's announcement that the technology would be embedded globally in its models. The code has since drawn more than 20,000 bookmarks on X and attracted over 100 contributors, with many incorporating the technology into their own projects.
Anthropic adopted the watermarking system to comply with the European Union's AI Act, which came into effect earlier this month. The new rules require model providers to label synthetic audio, image, video, or text in machine-readable form or face fines of up to 3 percent of annual turnover. The rules prohibit providers from marketing circumvention tools, though independent tools face no legal restriction.
Meyer told WIRED that some developers reject the watermarking principle, while others, including himself, pursue it as a technical challenge. Freelance content writers and social media creators have also asked him for assistance using the code, he said. The watermarking technique, called SynthID, was developed by Google and has been used by that company since 2023 to mark its AI-generated content.
Meyer's method uses non-watermarking language models to generate multiple rewrites with synonyms and reorganized content. The approach assumes other models will not insert watermarks, though 190 organizations—including OpenAI, Microsoft, and Meta—have signed the EU's transparency code of practice. New models must include watermarks from August, and existing models must integrate them by December.
Wayne Pan, chief technology and cofounder of Silicon Valley-based Haimaker, incorporated Meyer's open-source tool into his platform. Pan said he disliked the concept of Claude watermarking lightly edited content while remaining invisible to the user. Understanding SynthID's basic text approach makes developers "fairly sure the method works," Pan said, though Anthropic has not released its watermark detection software.
