Anthropic has announced that it will introduce digital watermarking for text and files generated by its AI models, including "Claude." This move aims to enable the automatic identification of content provenance in accordance with the "codes of practice on transparency" based on the European AI Act, which came into effect in August.
The watermarking applies to all Anthropic models released on or after August 2, with the open standard C2PA being used for files. Support is expected to expand to previously released models in the future. Since this technology is applied at the model level, it will also cover content created via API generation and various services such as Claude Code.
Source: Anthropic、Claudeなどの生成文に電子透かしを導入 欧州AI規則に対応 - Межа. Новини України. (Google News: Anthropic)