Anthropic has published a blog post explaining how its chatbot Claude's new watermarks will work, aiming to address user concerns and comply with the EU AI Act's Transparency Code. The watermarks will be undetectable to readers but detectable to those with a key, and will not impact the quality of Claude's output.
Key Insights
Anthropic will utilize the SynthID-Text approach for watermarking, ensuring that the watermarking process does not affect the quality of Claude's output. Additionally, while light editing may not completely remove the watermark, a complete rewrite will effectively eliminate it.
