tech
Claude will apply invisible watermarks to AI text and images
Posts from this topic will be added to your daily email digest and your homepage feed.
TL;DR
- Anthropic will mark AI-generated text and images with machine-readable data.
- This is to comply with new EU AI transparency regulations.
- Generated text will have embedded watermarks, and files will include digitally signed provenance metadata.
- The changes are invisible to humans but detectable by machines and platforms.
- New Claude models will mark AI content from day one; existing models are a work in progress.
- C2PA standard will be used for images, while text watermarking details are lighter.
- Watermarks are designed to persist even when content is copied or edited.
- Anthropic is developing tools for detecting these marks and metadata.
- The company acknowledges that these marking systems are not infallible.