Anthropic's Text Watermark Can Be Bypassed on Older Claude Models
Original headline: Anthropic's text watermark is easy to bypass on several older Claude models
Anthropic announced that it is adding text watermarks to four additional Claude models ahead of the EU AI Act deadline in December. However, on older versions such as Claude Opus 4.6 and Sonnet 4.6, API users can disable the watermark simply by changing a single sampling parameter. Anthropic responded to the loophole by issuing a compliance warning to users.
If watermarks are easy to bypass, the credibility of EU AI Act compliance verification is undermined.
Research angleHow does the ability to bypass text watermarks affect users' trust in AI-generated content? — Trust Theory
This story could become your next paper
HJITM works with researchers from study design through surveys, analysis and submission to international journals in generative AI and technology management. Bring us your research question.
This item is an AI-generated summary of reporting by The Decoder. Please check the original article for accuracy. Copyright in the original article belongs to The Decoder.