Anthropic to add invisible watermarks to all Claude-processed content

Anthropic will soon add invisible, machine-readable watermarks to text and other content processed by its Claude models, not just AI-generated output. The move aligns with the EU AI Act, which requires watermarking for AI-generated media, and applies globally to all new models from day one.

Text outputs will carry embedded watermarks, while other files will include signed provenance metadata where supported. Anthropic is going beyond EU requirements by watermarking even lightly edited content, such as grammar corrections.