Anthropic is starting to add machine-readable invisible watermarks to text generated by Claude. The feature covers Claude, Claude Code, Cowork, the API, and supported models on AWS, Google Cloud, and Microsoft Foundry. New models will ship with it from day one; existing models are being updated to support it.
The move is designed to satisfy the transparency requirements of the EU's AI Act, but Anthropic is rolling it out globally. The watermark is embedded directly into the text as it's generated — invisible to the naked eye, and it sticks around even after copy-pasting. Light editing, like changing a few words, may not be enough to strip it out.