Anthropic to Embed Invisible Watermarks Across Claude Line From August 2

From August 2, text generated by Claude will carry invisible watermarks. The measure is based on the transparency requirements of Article 50 of the EU AI Act, and Anthropic announced the plan in its help documentation. Violations are subject to penalties of up to 15 million euros or 3% of global annual revenue. The marks apply to all Claude products worldwide, not only EU users.

Anthropic has designed two layers of marking for AI-generated content. The first is a text watermark, which introduces an invisible statistical bias through token selection and is embedded in the text itself. The second is file provenance metadata following the C2PA standard, covering file types such as . svg, . png, and . jpg.

The marks cover all Claude product lines, including Claude Platform, Claude, Claude Code, Claude Cowork, and Claude Tag. Cloud customers accessing via AWS, Google Cloud, and Microsoft Foundry will also have the text watermark applied. August 2 is the effective date for new models, with a transition period for older models lasting until December 2.

Anthropic described the watermark's limitations in its documentation.

Anthropic is the first major AI company to deploy watermarks on text output at scale. The company has committed to providing watermark detection tools to users and third parties, though technical documentation has not yet been published. Paid users cannot disable the watermark, and API developers must assess their own compliance requirements.