Claude-generated content will be traceable even after copy-paste

Anthropic has confirmed that texts generated by its Claude chatbot will carry an invisible watermark, detectable even after copy-paste into emails or documents. The company is implementing this in response to Article 50 of the EU AI Act, which mandates that generative AI providers make their outputs machine-readable and detectable. Although the regulation is European, Anthropic plans to apply the watermark globally, including outside the EU. The watermark will be embedded in the text itself and will persist after copy-paste and even after modifications, affecting all Claude products including Claude Code, Claude Cowork, and Claude Tag, as well as use via AWS, Google Cloud, or Microsoft Foundry. For image files like PNG and JPEG, Anthropic will use C2PA metadata. However, the watermark only indicates that Claude was involved, not the extent of its contribution, so a human-written text corrected by Claude could be flagged, while heavily rewritten or short texts may lose the watermark. This has sparked criticism from users, though other providers like OpenAI, Google, and Meta will also need to comply with the AI Act, but they are not required to use the same method or apply it globally.