Anthropic has explained that its upcoming Claude text watermarking will use a version of Google DeepMind's open-source SynthID-Text approach. The system creates detectable patterns through word-choice probabilities rather than adding visible labels.
The company is adding text watermarks and C2PA support for Claude-processed images to meet transparency obligations under the European Union's AI Act, which requires machine-readable marking for synthetic audio, image, video, and text. The Verge reports that the feature is meant to make Claude-generated content detectable while remaining invisible to ordinary readers.
The trade-off is still contested. Text watermarks can be useful for compliance and provenance, but critics worry about robustness, possible effects on wording, and legal headaches when generated text is edited, translated, or mixed with human writing. The announcement shows that AI companies are moving from voluntary disclosure debates into concrete implementation choices shaped by regulation.