Summary
Anthropic announced it will implement invisible watermarks on all content processed by its AI models, not just generated, to comply with the EU's AI Act. This global rollout will apply to new models from day one, with text outputs carrying embedded watermarks and other files including digitally signed provenance metadata.