Anthropic Is Embedding Invisible Watermarks in All Claude Text

Anthropic is embedding invisible watermarks in all text Claude generates and attaching C2PA metadata to files (.png, .jpg, .svg). The labeling applies globally — to Claude, Claude Code, and the API — starting in August 2026: new models immediately, existing models through a transition period. The watermark does not affect text quality and "may survive some edits," but heavy editing or translation removes it.

This changes the rules for the entire industry: the provenance of AI-generated content can now be verified technically, not just by eye. Anthropic signed the EU AI Act Code of Practice on Transparency — European regulatory pressure is turning invisible labeling from an experiment into a production standard. Verification tools for third parties are planned, but no release date has been announced.