Anthropic now embeds invisible watermarks in all text generated by Claude. These watermarks persist when text is copied and apply across all platforms, including the API, Claude apps, Claude Code, Claude Cowork, Claude Tag, as well as versions accessible through AWS, Google Cloud, and Microsoft Foundry. Detection tools for users and third parties will be released, with a compliance deadline set for December 2, 2026 under the EU AI Act. A positive detection merely indicates that content may have been processed by Claude, without constituting proof of cheating since the model is also used for proofreading, translation, and summarization.
Source: Read the original article
