Policy + product
Anthropic watermark (Claude)
The Anthropic watermark story is really two systems on Claude: an embedded text watermark, and signed C2PA provenance on supported files—rolled out alongside EU AI Act transparency commitments for models launching on/after August 2, 2026.
What Anthropic publicly committed to
- Machine-readable marks in Claude-generated text
- Signed provenance metadata on supported files
- Detection support for users and third parties (docs forthcoming)
- Worldwide marking on supported models/products
What that means for you
If you ship Claude output into customer-facing docs, SEO pages, or support replies, the text may carry a detectable Claude signal. A Claude watermark remover workflow helps when you need meaning-preserving neutralization before reuse.
Not affiliated with Anthropic.