Anthropic’s Claude to mark all AI content, including text

Anthropic, a leading AI company, has announced that its Claude models will include identity markings to comply with European Union regulations. This move is in line with Article 50(2) of the EU AI Act's Code of Practice on Transparency of AI-Generated Content. The new requirement will come into effect for Claude models launched in the EU on or after August 2.

The AI-generated text produced by Claude models will carry embedded watermarks, ensuring that the content is easily identifiable as AI-generated. This watermark will be imperceptible to human readers and will remain even if the text is copied, pasted, or lightly edited. Moreover, when Claude generates supported file types such as .svg, .png, or .jpg, digitally signed provenance metadata will be attached to the file.

However, it is essential to note that the marks are not foolproof. Even if a mark is detected, it does not guarantee that Claude was the original author of the content. The company also warns about potential false negatives due to excessive editing, short text lengths, or stripped metadata.

These AI content markings will be applied to outputs generated by supported Claude models across various platforms, including Claude Platform (API), Claude, Claude Code, Claude Cowork, and Claude Tag. This includes applications accessed through AWS, Google Cloud, or Microsoft Foundry. Anthropic has promised to provide more information about how it will support users and third parties in detecting Claude's marks.

Anthropic's initiative aims to enhance transparency and provide users with useful context about the origin of the information they consume. As AI-generated content becomes more commonplace, greater transparency can help people make informed decisions about the content they interact with. The company is committed to complying with its legal obligations while working towards a more transparent AI landscape.