Anthropic to embed invisible watermarks in all Claude text outputs globally

The Decoder · rss · 2026-08-11

Anthropic will embed invisible watermarks in all text generated by Claude models and sign files using the C2PA standard. The policy applies globally, and the marks are designed to persist even after some editing.

New models shipping from August 2026 onward will have labeling built-in from day one. Anthropic also plans to provide detection tools to enable third-party verification.

Original post →

More from Models

Models channel →