Anthropic Adds Invisible Watermarks and C2PA Signatures to Claude

As generative AI spreads across publishing, education and software development, identifying who created a piece of content — and how it was altered — has become a pressing policy issue. Anthropic is adding machine-readable provenance signals to Claude output as part of its response to Article 50 of the European Union’s AI Act, whose transparency requirements took effect on Aug. 2, 2026. The measures are intended to help platforms and verification services distinguish AI-assisted material from unmarked content.

Anthropic said Claude models launched on or after Aug. 2, 2026, embed an imperceptible watermark directly in generated text, while supported PNG, JPG and SVG files carry digitally signed provenance metadata based on the C2PA standard. Earlier models are due to receive marking capabilities during a transition period. The text signal can survive copying, pasting and some editing without changing readability, but extensive rewriting may defeat detection. File credentials can likewise be stripped, making the system a provenance indicator rather than definitive proof of authorship.