Anthropic has announced significant updates to its documentation concerning the labeling of AI-generated content from its Claude models. Starting August 2, 2026, new models introduced in the European Union will feature invisible watermarks and digital certificates of origin embedded within the text, in compliance with the EU's AI regulations and transparency standards.
This labeling initiative will also apply to Claude products globally, including those distributed via AWS, Google Cloud, and Microsoft Foundry.
For models released before the August 2026 deadline, a four-month grace period will be in effect. Anthropic plans to introduce verification tools for these labels in the near future. However, the company has warned that watermarks may not always be reliable, as they can become less detectable after editing, translating, or if the file's metadata is stripped away.
