WASHINGTON, August 12 (WAM), Anthropic announced the launch of a new tool that helps detect AI-written narratives by embedding an imperceptible watermark in texts generated by Claude models.

The company explained that the watermark does not affect the meaning or readability of the text and is transferred when copied and pasted, noting that it will be applied to content produced by Claude worldwide, including when used through cloud service providers, with plans to provide detection tools for external parties.

Claude models launched this month or later will support this feature from the start, while Anthropic works on adding it to older models, a move it said is part of its commitment to enhancing transparency under the EU AI Act.

The watermark provides publishers and educational institutions with an additional means of verifying the use of AI in writing at a time when the publishing industry faces increasing challenges regarding AI-generated content.

Anthropic warned that extensive editing, rewriting, translation, or merging Claude outputs with other writings may make the watermark undetectable. It also does not conclusively prove that Claude was the original creator of the text.

- Khla -