Claude Text Watermark and Detection API Access (EU AI Act)

Coffee Summary

• FACT: Anthropic is rolling out SynthID-Text-style watermarking so Claude-generated text can be statistically checked for Claude involvement (EU AI Act / Code of Practice transparency).
• Watermarking changes the source of randomness in low-stakes token choices — no hidden characters, no per-user tracing, negligible quality impact per Anthropic.
• Detection API is in private preview for eligible orgs (regulators, law enforcement, media, fact-checkers, researchers, educational orgs, EU civil society, and obligated enterprises).
• Limits: not definitive authorship; weak on short / heavily rewritten text; proofreading may leave too little signal.
• Files use separate C2PA content credentials — different from text watermarks.