Anthropic has announced plans to release a watermark detection API that will enable third parties to verify whether a piece of text was generated by its AI model, Claude. According to The Decoder, this new tool is expected to become available soon and is designed to help identify AI-produced content more reliably.
The technology behind the API builds upon Google's SynthID method, adjusting the randomness during word selection in the text generation process. This technique aims to embed a detectable watermark without compromising the quality of the generated text. However, The Decoder notes that the approach may face challenges with fact-heavy content, programming code, or heavily rewritten material.
For Japanese markets, where AI-driven content and automation are increasingly integrated into sectors such as FX and equities analysis, tools that can authenticate AI-generated text could become important for maintaining transparency and trust in digital communications.