tech π€ Anthropic plans invisible watermarking for Claude text
Anthropic is implementing a system to watermark AI-generated text to help identify AI content globally. This is partly to comply with the EU's Code of Practice, with initial application worldwide. The watermark works by changing the source of randomness during text generation, not by adding visible characters. Internal tests show this has no negative impact on Claude's quality or readability. Future models will carry this mark, though models before August 2, 2026, are being updated. π€