AI Watermark Removers Flood Web Days After Claude Marks Text for EU Act, Drawing 4,500 Stars
Updated
Updated · BleepingComputer · Aug 14
AI Watermark Removers Flood Web Days After Claude Marks Text for EU Act, Drawing 4,500 Stars
3 articles · Updated · BleepingComputer · Aug 14
Summary
Days after Anthropic enabled invisible Claude text watermarks for EU AI Act compliance, remover tools spread across GitHub and new websites, with one project alone topping 4,500 stars and a related post drawing more than 2 million views.
Anthropic has not published its watermarking method or a public detector, so most services cannot prove they remove the actual mark rather than just hidden characters or file metadata.
The verifiable part is limited: tools can strip zero-width characters and C2PA, EXIF or XMP metadata, but Anthropic says the text watermark is embedded in word choice and would require heavy rewriting to disrupt.
Independent code review already found gaps, while some commercial sites promise 'undetectable' output using scores from ordinary AI detectors instead of Anthropic's unpublished watermark detector.
Anthropic switched on marking for models launched on or after Aug. 2 to meet EU AI Act Article 50, which carries penalties of up to 15 million euros or 3% of global turnover.