rvachev.orgEN / RU / 🤖
← Back to essays
· Essay · 1 min

Claude Begins Watermarking Text

Claude has started placing invisible watermarks on all generated text.

🔍 Claude has started placing invisible watermarks on all generated text.

Anthropic introduced this on August 2, 2026, as a response to Article 50 of the EU AI Act, which requires disclosure that content is AI-generated. The watermark works globally, not just for European users, and covers claude.ai, API, Claude Code, Cowork, Tag, as well as access through AWS, Google Cloud, and Microsoft Foundry.

https://techcrunch.com/2026/08/11/anthropic-says-it-will-watermark-text-generated-by-its-ai-models/

The watermark is invisible during normal reading, survives copy-paste, and is partially retained after editing. For files, instead of a watermark, Anthropic uses signed provenance metadata according to the C2PA standard.

Anthropic promises a separate public API to check text for such marks - for example, it will be possible to verify if a PR was written by Claude Code. However, the company itself admits: the mark only indicates that the text went through Claude, not who authored it or if it was changed afterward. Other labs are developing similar marking schemes.

#anthropic@rvnikita_blog #claude@rvnikita_blog #ai@rvnikita_blog #regulation@rvnikita_blog