Anthropic Adds Invisible Watermarks to Claude Text Worldwide, Japan Included
# watermarking# C2PA# EU AI Act# generative AI
Key Points
- Anthropic will watermark text generated by Claude and attach C2PA-compliant signed metadata to supported files
- The move follows transparency obligations under the EU AI Act, applied from 2 August
- It covers all models and services worldwide, Japan included
Analysis
The story is not the watermark but the spillover: a rule written in one jurisdiction becomes a global product specification. You do not have to be in the EU for it to reach the tool on your desk. Practically, traceability is what changes. Invisible marks in text and signed file metadata make it technically checkable whether something was AI-generated and whether it was altered afterwards. Publishers, academics and ghost-writing workflows should settle their disclosure practices now.