Claude Now Embeds Invisible Watermarks in All Generated Text

Starting August 2, Claude models weave an imperceptible watermark into generated text that survives copy-paste and light edits, and attach signed C2PA provenance metadata to supported file types — worldwide, not just in the EU.

Anthropic rolled out two complementary provenance mechanisms in response to the EU AI Act's Article 50 transparency requirements, but chose to apply them globally:

• Text watermarking: an invisible signal baked into generated text at the model level. Users see nothing different; the signal persists through copy-paste and partial edits. Detection requires Anthropic's verification tooling, not available to end users directly.
• File provenance (C2PA): generated SVG, PNG, and JPG files receive signed Coalition for Content Provenance and Authenticity metadata — the same open standard used by Adobe Firefly, OpenAI image tools, and Google Imagen.

Coverage applies across Claude Platform (API), Claude, Claude Code, Claude Cowork, and Claude Tag.

Why it matters: If you're building a product on Claude, your users' output is now watermarked. This is largely invisible in practice, but worth knowing when: (1) you're integrating content-verification pipelines that check for AI-generated text; (2) your users paste Claude output into third-party tools that scan for watermarks; or (3) you're in a domain (journalism, legal, compliance) where AI provenance is actively scrutinized.

Sources: How Claude marks AI-generated content — Claude Help Center · Anthropic embeds invisible watermarks globally — Crypto Briefing · Claude invisible watermarks — Interesting Engineering