Anthropic shares more details about how Claude's new watermarks will work

Anthropic says it will add watermarks to text generated by its Claude chatbot as part of compliance with the EU AI Act’s Transparency Code. In a blog post published Friday, the company addressed questions including how the watermark works, whether it can be hidden through editing, and how it affects code. Anthropic said watermarked outputs are indistinguishable from unwatermarked ones to readers, using the SynthID-Text approach developed by Google DeepMind in 2024, and it plans to release a watermark detection API. The company acknowledged users can rewrite text to try to remove watermarks: light editing likely won’t eliminate them, while a complete rewrite could. It also said detection in edited text depends on length and how heavily Claude modified it, and that code should carry less watermark due to fewer equivalent word choices.