Anthropic Adds Invisible Watermarks to AI-Generated Content

Anthropic is embedding a "transparent marker" into content created by its AI model Claude, making it machine-readable. This invisible watermark will allow the AI to identify its own output, even if the text is copied or slightly altered.

The company announced on its official website that this feature aims to ensure the traceability of AI-generated content. The system will apply a hidden watermark to text, enabling detection of its AI origin through subsequent modifications. For files, Anthropic is incorporating digital signatures that indicate the source of the content.

This initiative is a direct response to the European Union's AI Act, specifically its "Transparency Regulation" for AI-generated content, which took effect on February 2.

The regulation requires clear labeling of AI-generated text and images to prevent them from being mistaken for human creations or being used for disinformation and deepfakes. While implemented to comply with EU regulations, the marking feature will be available to all Claude users globally. New models released after February 2 will include this functionality from launch, with existing models to receive it sequentially.