Anthropic adds machine-readable watermarks to Claude globally

Anthropic rolls out machine-readable watermarks for Claude globally

Anthropic adds invisible metadata to texts generated by new Claude models. The initiative covers all platforms including AWS, Google Cloud, and Microsoft Foundry, but watermarks are not a binary proof of authorship.

Author: Гика • PC •

Anthropic has announced the global rollout of machine-readable watermarks for content generated by its Claude models. The update applies across all interfaces, including integrations with AWS, Google Cloud, and Microsoft Foundry. While driven by compliance with the EU AI Act, Anthropic intends to implement this transparency measure worldwide.

New Claude models launching on August 2, 2026, will embed these watermarks directly at the model level. This ensures that the metadata persists even when content is copied and pasted into external websites or documents.

Functionality and limitations

The watermarks are invisible to users and do not affect text quality or readability. They serve as a signal that the content was processed by Claude, rather than written entirely by it. Anthropic notes that while the watermark is robust against minor edits, extensive paraphrasing, translation, or heavy rewriting can obscure or remove the signature.

Integration with C2PA

This feature complements existing signed provenance metadata based on the C2PA standard (Coalition for Content Provenance and Authenticity), which is used for file types like SVG, PNG, and JPG. Together, these tools help verify the origin of files and detect potential tampering.

Future availability

Beyond Anthropic's own products (Claude Code, Claude Cowork), the watermarking will be available on third-party cloud platforms that support the integration. Anthropic plans to release tools for third parties to verify these watermarks in the future, though full technical specifications are not yet public.

Гика

We use cookies