Anthropic to Embed Invisible Watermarks in Claude-Generated Text Globally

socialwithaayan · x · 2026-08-11

Anthropic confirmed it will comply with the EU AI Act's transparency requirements by embedding invisible watermarks in text generated by new Claude models starting August 2, worldwide.

The watermark is invisible, persists through copying, and may survive light editing. However, this raises concerns about false positives: users who only employ AI to fix typos might be flagged identically to those who generate entire texts via AI. Furthermore, the watermark disappears if the text is heavily rewritten, meaning honest users get caught while those deliberately hiding AI usage evade detection.

Related event: Claude Rolls Out Invisible Watermarks and Metadata, Sparking Compliance and Tech Debates(115 posts)→

Original post →

More from Models

Models channel →