Anthropic implements watermarking for Claude AI content

Anthropic has implemented invisible text watermarking and digitally signed provenance metadata for content generated by its Claude AI models. This initiative aims to increase transparency regarding AI involvement, aligning with European AI transparency requirements. The text watermarking is applied at the model level and is designed to remain embedded even if content is copied and pasted, without affecting the quality, length, or readability of the output.

To complement this, Anthropic has launched a content checker tool at claude.com/check-content. Currently, this tool is limited to multimedia files—such as JPEGs, GIFs, MP3s, and MOVs—using the Coalition for Content Provenance and Authenticity (C2PA) standard. While the public tool does not yet support text files, Anthropic is rolling out a private Detection API for text watermarks to eligible organizations as required by EU law. This API is intended to help businesses and publishers identify Claude-generated text.