How Claude's text watermarking works
Anthropic is preparing to release an API that allows third parties to detect if text was generated or edited by Claude. The system utilizes Google’s SynthID technology to embed an invisible watermark by subtly adjusting word selection patterns during generation. This approach ensures the mark persists even when human users modify the output. The tool addresses the growing challenge of identifying machine-generated content, though Anthropic notes that the detection method has limitations when applied to factual or short-form writing.
Covered by 5 sources · 6 articles
- AAnthropic↗Aug 14
- TThe Decoder↗Matthias BastianAug 14
- AArs Technica↗Ashley BelangerAug 13
- HHacker News↗pseudolusAug 13
- HHacker News↗padolseyAug 13
- BBleepingComputer↗Aug 14