Google subsidiary DeepMind has introduced SynthID-Text, a system for watermarking text generated by large language models (LLMs). By subtly altering word probabilities during text generation, SynthID-Text embeds a detectable statistical signature without degrading the quality, accuracy, or speed of the output, as described by Pushmeet Kohli and colleagues in the journal Nature. While not foolproof, as edited or summarized text can obscure the watermark, the tool has been successfully tested on 20 million prompts and integrated into Google’s Gemini chatbot.
Although an important step forward, SynthID-Text is not yet a comprehensive solution for identifying AI-generated content, as IEEE Spectrum reports. Google has made the tool available to some developers and businesses, but widespread adoption and interoperability among AI companies would be necessary to effectively combat the flood of AI-generated text online. Bruce MacCormack of the Coalition for Content Provenance and Authenticity (C2PA) sees promise in the approach but cautions that practical implementation challenges remain, describing the research as “the first step on a long road.”