Anthropic has revealed more details on how it plans to watermark Claude's responses, ensuring compliance with EU regulations. The company explains that the watermark is invisible to the human eye but detectable by those holding the key.

The SynthID-Text approach will be used, and a watermark detection API will be released soon. While some users fear this might interfere with their experience, Anthropic reassures that 'watermarking does not impact the quality of Claude’s output.'

However, editing the text can sometimes remove the watermark, though a complete rewrite would likely nullify it. Code generated by Claude will have less visible watermarks due to its functional nature.

The move is part of a broader initiative, with other major model developers implementing similar measures to ensure transparency in AI-generated content.