Anthropic has introduced invisible, machine-readable watermarks for AI-generated content created with Claude. The hidden markers are designed to remain invisible to readers while helping identify text generated by Claude without changing how it appears. The move is linked to Anthropic’s transparency commitments under the EU AI Act, with marking supported in the EU from August 2 and planned worldwide. The system also supports provenance data for images, including C2PA metadata that can indicate whether a file was processed by Claude and help detect tampering.
