Anthropic is actively broadening its text watermarking initiative to include four additional Claude models, aiming to comply with the EU AI Act's impending December deadline. However, it has come to light that the older models, Claude Opus 4.6 and Sonnet 4.6, present a significant loophole. Developers can circumvent the watermark by merely adjusting a single sampling parameter. While Anthropic has acknowledged this issue, the response has been limited to a compliance warning rather than a robust solution. This situation raises important questions regarding the effectiveness of regulatory measures and the potential risks associated with AI deployment. A thorough evaluation of the implications for stakeholders is essential as the landscape of AI regulation continues to evolve.