Developers launch open-source tools to strip Anthropic’s AI watermarks amid EU push

Anthropic introduced an imperceptible watermark that embeds a statistical pattern in text produced by specific Claude models, a step taken to meet the European Union’s AI Act requirements. Critics warn the label could persist even after light editing, leading some users to cancel subscriptions. In response, Google Trends showed a 60% week-on-week rise in searches for “AI watermark remover” in the United States, and developers released open-source projects like Guillaume Meyer’s “Watermarks Remover,” Sabrina Ramonov’s browser-based cleaner, and Ansh Aneja’s Claude-focused MarkScrub.

These tools remove hidden characters, metadata, or rewrite text to disrupt the watermark, though they do not guarantee removal and their effectiveness remains unverified. Researchers note that watermarking can always be circumvented, and EU guidance obliges providers to make markings resilient but does not forbid third-party removal, leaving a legal gray zone. Anthropic’s policy prohibits presenting AI-generated output as human-written, a rule that may gain importance when the company later launches a text-detection API.