Anthropic Details Claude AI's Invisible Text Watermarking Method
Anthropic has detailed how its Claude AI will implement invisible text watermarks to comply with EU's AI transparency regulations.
Artificial intelligence company Anthropic has clarified the method it will use to embed invisible watermarks within text generated by its Claude AI. This initiative aims to meet the transparency requirements mandated by the European Union's AI Act, which concerns the identification of synthetic content.
Anthropic announced on Friday that Claude's text marking system utilizes a "version of the SynthID-Text approach." This open-source watermarking technology, developed by Google DeepMind, creates detectable patterns by analyzing word probabilities within the generated text.
In addition to text watermarking, Claude will also support the C2PA standard for image processing. These features are being introduced to fulfill Anthropic's obligations under the EU AI Act. The regulation mandates that synthetic audio, image, video, and text must include machine-readable marks.
The technology is designed to allow users and regulators to distinguish AI-generated content from human-created material. The watermarks are intended to be imperceptible during normal reading and not affect text quality, yet be detectable through specialized analysis tools.