Anthropic has begun adding invisible digital markers to texts generated by its Claude neural networks. This decision comes in response to the requirements of the European Union's Artificial Intelligence Act, as reported by infohub.kz.
According to Anthropic, the invisible watermarks are embedded directly into the text generated by Claude models released after August 2, 2026. The markers do not affect readability, meaning, or quality of responses, but they persist through copying and partially withstand editing.
The company has taken this step to comply with Article 50 of the EU AI Act, which mandates transparency for synthetic content. The labeling applies to all of the developer's services worldwide, including the Claude interface, API, the Claude Code tool, and models on cloud partners' platforms.
For media files, Anthropic uses the C2PA digital signature standard, while for text it has developed a built-in watermark. The company emphasizes that the presence of the mark is not absolute proof of AI authorship, as users often employ the model merely for translation or proofreading of existing texts.