Anthropic Embeds Invisible Watermarks in Claude AI Text to Meet EU Transparency Rules

Anthropic began embedding machine-readable watermarks in text generated by new Claude models starting August 2, 2026, in response to transparency requirements under the EU AI Act.

The system applies globally, not only in the European Union. It uses two methods: invisible signals woven directly into generated text, and digitally signed C2PA provenance metadata attached to supported file types such as .svg, .png, and .jpg.

The text watermark is integrated at the model level, meaning it is present across all surfaces where Claude is used, including the Claude web interface, API, Claude Code, and third-party cloud partners such as AWS, Google Cloud, and Microsoft Foundry. The signal is designed to persist through copying and pasting and may survive some forms of editing.

Anthropic said the marks do not serve as definitive proof of AI authorship. A detected mark indicates only that content may have been processed by Claude, which could include tasks like proofreading, translating, or summarizing human-written work. The absence of a mark does not guarantee content is human-generated, as the signal can be weakened through heavy paraphrasing, translation, or format conversion.

Models released before August 2, 2026, are not immediately equipped with the marking capabilities. Anthropic said it is working to incorporate them during the transition period allowed by the EU AI Act.

The company plans to release technical documentation and tools to help third parties detect the watermarks. The move follows similar transparency requirements that took effect under the EU AI Act's Article 50 in August 2026.