Claude’s text output is about to carry a hidden mark of origin. Anthropic has confirmed, via an updated support page, that it will watermark the writing its models produce, a step taken to meet Europe’s AI Act transparency requirements that became enforceable on August 2.
Models released in the EU on or after that date will embed the watermark automatically at launch. Generated files get an additional layer of protection through digitally signed provenance metadata built on the C2PA open standard. Older models will get the capability as well. Anthropic says the mark is part of the text itself, so it rides along through copy and paste and can survive light editing.
Because the watermark is applied at the model level, it appears regardless of which surface generated the content, covering the Claude platform API, Claude itself, Claude Code, Claude Cowork, and Claude Tag. Anthropic has not clarified how much editing would strip the mark.
The announcement puts Anthropic in a growing club. Black Forest Labs, Google, Meta, Microsoft, OpenAI, and Synthesia have all signed onto the EU code, and the wave extends beyond Europe – Suno said last week it would watermark AI songs amid legal fights, and Substack has begun flagging AI-generated newsletters.
