Anthropic Develops Invisible Watermark for Claude Text

Anthropic is adding invisible watermarks to Claude's text and output files to comply with the EU AI Act, detectable even after copying or editing. However, education experts say the technology cannot fundamentally stop students from cheating with AI.

2026-08-20 ~ 2026-08-21 · 3 related posts

• Anthropic's Claude watermark fails to stop cheating — adariostrange · 2026-08-20
• Anthropic is building a Claude text watermark that survives copy, paste and edits — Matt Wolfe · 2026-08-20
• How Claude's text watermark works: modifying token selection with a keyed function — blaizedsouza · 2026-08-21