Anthropic outlines marking system for AI-generated content under EU regulations
Anthropic has committed to marking AI-generated content in compliance with the EU AI Act's Code of Practice. Starting August 2, 2026, new Claude models will feature embedded watermarks and provenance metadata to enhance transparency. Existing models will also receive updates to support these marking features during a transition period.
%22%2F%3E%3Crect%20width%3D%22800%22%20height%3D%22450%22%20fill%3D%22url(%23glow)%22%2F%3E%3Ctext%20x%3D%22400%22%20y%3D%22275%22%20text-anchor%3D%22middle%22%20fill%3D%22%2360a5fa%22%20font-family%3D%22system-ui%2C-apple-system%2Csans-serif%22%20font-size%3D%22132%22%20font-weight%3D%22800%22%20letter-spacing%3D%22-6%22%20opacity%3D%220.9%22%3Ecn%3C%2Ftext%3E%3Ctext%20x%3D%22400%22%20y%3D%22318%22%20text-anchor%3D%22middle%22%20fill%3D%22%2360a5fa%22%20font-family%3D%22system-ui%2C-apple-system%2Csans-serif%22%20font-size%3D%2211%22%20font-weight%3D%22600%22%20letter-spacing%3D%223%22%20opacity%3D%220.38%22%3ECONTEXT%20NEWS%3C%2Ftext%3E%3C%2Fsvg%3E)
Anthropic has signed the EU AI Act's Article 50(2) Code of Practice on Transparency of AI-Generated Content.
New Claude models launched in the EU on or after August 2, 2026, will include machine-readable marking at launch, featuring embedded watermarks and digitally signed provenance metadata.
Marking will be applicable across all platforms where Claude is offered, including Claude Platform (API), Claude Code, Claude Cowork, and Claude Tag.
Existing Claude models will receive updates to support marking features during a transition period before August 2, 2026.
The marking system uses two techniques: embedded watermarks in text and signed provenance metadata for generated files, following the Coalition for Content Provenance and Authenticity (C2PA) standard.
Detection mechanisms will be developed to help users identify Claude's marks, although a detected mark does not confirm full content provenance.
Limitations of the marking system include the possibility of content modification after processing and the absence of a mark not necessarily indicating non-AI generation.