Anthropic has expanded its text watermarking technology to four additional Claude AI models as part of efforts to comply with the EU AI Act’s upcoming December deadline, according to The Decoder. This move aims to enhance transparency and traceability of AI-generated content across its product lineup.
However, The Decoder reports a loophole remains on two older models, Claude Opus 4.6 and Sonnet 4.6, where developers can circumvent the watermark by adjusting a single sampling parameter. Rather than patching the vulnerability immediately, Anthropic has issued a compliance warning addressing the issue.
For Japanese markets, where regulatory scrutiny on AI and digital content is increasing, Anthropic’s approach highlights the challenges global AI firms face in balancing compliance with product flexibility, especially as similar regulations are anticipated in the region.