How Claude Marks AI-Generated Content

Anthropic has announced its plan to mark content generated by Claude in a machine-readable manner in accordance with the Transparency Code of Practice under Article 50(2) of the EU AI Act.

• New models released in the EU after August 2, 2026 will support marking capabilities from the time of release.
• Marking will be applied to outputs from supported models across products and global services, including the Claude Platform (API), Claude, Claude Code, Claude Cowork, and Claude Tag.
• Work is underway to add marking capabilities to existing models released before August 2, 2026.

Claude marks content in two ways.

• Text watermark: An invisible watermark is inserted into generated text. The watermark is applied at the model level and may persist after copy-pasting and through some editing.
• Signed provenance metadata: Signed provenance metadata based on the Coalition for Content Provenance and Authenticity (C2PA) standard is attached to supported files such as SVG, PNG, and JPG. This metadata allows verification of whether Claude processed the file and whether it has been tampered with.

Anthropic plans to provide methods for users to detect Claude's text watermarks and file provenance metadata, with specific detection mechanisms to be disclosed in future technical documentation. However, due to platform-specific feature differences, signed metadata may not be supported in cloud partner environments such as AWS, Google Cloud, and Microsoft Foundry.