Claude now watermarks the text and files it generates, worldwide
Models from Aug 2 on weave an invisible mark into generated text and sign files with C2PA — EU-driven, applied globally, and your compliance call to make.
Copy markdown
Your Claude output now carries an invisible mark
Text from any Claude model released Aug 2 or later now has a watermark woven into the words themselves — it survives copy-paste and some editing. It's live across the Claude Platform API, Claude Code, Cowork, and the app, not just the consumer chat.
Files get C2PA provenance, not just prose
Generated SVG, PNG, and JPG files carry signed C2PA provenance metadata that flags tampering. The design assets and images you export are marked too, using the same open standard Adobe and others already read.
Why now: the EU AI Act's transparency code
The EU AI Act Transparency Code took effect Aug 2, requiring AI-generated content to be machine-detectable. Anthropic is applying the marks worldwide rather than EU-only, and retrofitting older models over time.
What it means if you ship Claude output
If you deploy Claude-generated text or files in a product, you own your EU Article 50 transparency obligations — assess before you deliver. And treat the mark as a signal, not proof: heavy editing or unsupported formats can strip it, and its presence doesn't prove Claude wrote the whole thing.
Elsewhere: Claude Code 2.1.231 fixes Slack MCP login
Today's Claude Code release fixes MCP OAuth sign-in failing with a redirect-URI mismatch for servers using pre-registered clients like Slack. Update if your Slack connector login had been breaking.