Anthropic has updated its documentation regarding the labeling of AI-generated content associated with its Claude models. Starting from August 2, 2026, all new models launched in the European Union will incorporate invisible watermarks and digital certificates of origin directly into the text, as mandated by the EU's AI legislation and transparency code.
This initiative will extend to Claude products worldwide, including versions available on AWS, Google Cloud, and Microsoft Foundry.
For existing models released prior to August 2, a four-month grace period has been granted, and tools for verifying the labels will be announced later. However, Anthropic has cautioned that watermarks do not provide a foolproof solution, as they can be harder to detect after editing, translating, in brief texts, or if metadata is removed from the file.