Identifying AI-generated content: SynthID from Google DeepMind

A short note on something that matters more than its size suggests. Google DeepMind has made its SynthID watermarking technology generally available, letting developers watermark and detect text written by generative AI models.

SynthID's watermarking technique is imperceptible to humans but detectable for identification.

Others, including OpenAI, have been working on the same problem but have cited various reasons for delaying release. SynthID is available for text, images, video and music in beta — a small but important step toward provenance and authenticity as synthetic media becomes ubiquitous.