Anthropic Brings SynthID-Text Watermarks to Claude for EU AI Act Compliance
Anthropic reveals how it plans to add invisible watermarks to text generated by Claude, using a version of SynthID-Text, an open-source watermarking technology developed by Google DeepMind. The system creates detectable patterns by subtly influencing which words the model selects during generation. The feature, alongside C2PA support for Claude-processed images, helps Anthropic comply with the European Union's AI Act.
The EU AI Act requires synthetic audio, image, video, and text to carry machine-readable marks so the content can be identified as artificially generated or manipulated. Anthropic explains the approach with a simple example: given the sentence "The weather today was cold and…", the next word is unlikely to be "sugary" but could plausibly be "overcast" or "grey." When multiple word choices carry essentially the same meaning, the watermark can steer the selection in ways that are invisible to readers.
Anthropic says the watermarks will not increase costs for users and will not have any practical impact on the quality or content of Claude's outputs. The company positions the change as a transparency measure, allowing AI-generated text to be detected as machine-made while preserving the natural reading experience for humans.