Anthropic Adds Watermarks and C2PA Metadata to AI-Generated Content

AI firm Anthropic has begun embedding watermarks into text responses and Coalition for Content Provenance and Authenticity (C2PA) metadata into media files generated by its Claude models. These measures aim to increase the transparency and traceability of AI-generated content.

The watermark for text is designed to travel with the content, persisting through copying, pasting, and some editing. This watermarking is applied at the model level, ensuring its presence regardless of the Claude product or interface used.

For media files like .svg, .png, or .jpg, Anthropic is adding metadata compliant with the C2PA open standard. This industry-standard format records information about content provenance. A signed metadata label indicates that a file was processed by Claude and can help detect tampering.

These implementations align with Anthropic's compliance efforts for the EU AI Act, specifically Article 50 concerning transparency of AI-generated content, which came into effect globally. Claude models launched in the EU on or after August 2, 2026, will support machine-readable marking at launch.

Anthropic intends to release detection tools and technical documentation soon, allowing users and third parties to identify the embedded watermarks and metadata. The company acknowledges ongoing challenges in preventing circumvention, such as metadata stripping via screenshots or encrypted platforms.