Etzioni on AI: Claude is marking its text — Caveat Promptor!
Anthropic embeds invisible watermarks in Claude output. Users can't see them, can't disable them, and they persist across copy-paste.
Why it matters
A significant transparency and user control issue: Claude now marks all text invisibly, raising questions about consent, detectability, and what hidden metadata means for enterprises deploying Claude at scale.
The key facts
10 to knowInvisible watermark embedded in Claude output launched Aug 2 or later
Watermark persists when text is copied and pasted
Users cannot opt out or disable the watermark
Watermark is not visible to end users
Raises transparency and consent questions for enterprise deployments
Claude models launched Aug 2, 2026 or later embed invisible watermarks
Watermark persists through copy-paste operations
No user opt-in, no visibility, no toggle to disable
Watermark is embedded in text itself, not metadata
Byline: Oren Etzioni (expert commentary on AI practice)
Go to the source
GeekWiregeekwire.com
Publisher excerpt: Claude models launched on or after Aug. 2 embed an invisible mark in everything they write. It's woven into the text itself, so it travels when you copy and paste. You didn't opt in, you can't see it, and you can't turn it off. Here's how it works, and what it means.