Anthropic implements machine-readable watermarks on all processed content to comply with EU AI Act

Anthropic announced implementation of machine-readable watermarks on content processed by its models to ensure compliance with the European Union's AI Act. The watermarks, which remain invisible to users, will be embedded in text outputs and included as digitally signed metadata in other media formats. Although the EU requirement applies only to content generated after August 2, Anthropic will retroactively update previously released models by December 2026. The company is deploying watermarks globally to all new models from launch, exceeding regulatory requirements that exclude certain assistive functions like grammar correction.