Anthropic has begun adding invisible, machine-readable watermarks to Claude-generated content to comply with the European Union’s AI Act, but developers say they found ways around the system within hours. Claude’s text watermarking uses Google’s SynthID technology, subtly influencing word and phrase choices to create patterns that detection software can recognize without changing how the text appears to readers.

Developer Guillaume Meyer released an open-source method that attempts to remove the watermark by having another AI model rewrite the text with synonyms and altered sentence structures. Critics argue invisible watermarking could produce false accusations of AI use, particularly when someone merely uses AI to edit or polish their own writing, while Anthropic says its detection only indicates a probability that Claude was involved. The larger problem may be an escalating technological cat-and-mouse game: the EU is requiring AI providers to make synthetic content detectable, while independent developers remain free to create tools designed to defeat those detection methods.

Source: Coders Say They Already Found Workarounds to Claude’s Invisible Watermarks

I know it's not exactly the same, but I generated an image with Gemini and when I asked Claude to work on it, it removed Gemini's AI watermark without me even asking. Which, ok, that's cool. But not surprising Claude's watermark can so easily be stripped off as well.