Claude to embed invisible watermarks on AI outputs for transparency

Anthropic says Claude-generated text will carry imperceptible, embedded watermarks and Claude-generated images will include provenance data via the C2PA standard to align with EU AI Act transparency rules. The marks are model-level and travel with content across surfaces, with detection tools to be released later. Implementation is a future commitment with a grace period for existing products, and there are caveats about watermark robustness and the possibility of marks being stripped by platforms.