Summary
How can you tell whether an image, a video clip or even a DNA sequence was made by AI? This episode of the Google DeepMind podcast, hosted by Professor Hannah Fry, explores watermarking as an answer, with guests Pushmeet Kohli, VP of science at Google DeepMind and one of the architects of SynthID, and biosecurity researcher Jeremy Radcliffe.
The guests outline what makes a good watermark: it should be imperceptible to people, hard to remove even after edits or transformations, and easy to integrate and detect at scale. Kohli describes how SynthID’s embedder and detector are trained together against an adversary that crops, rotates, shrinks or adds noise to content, so the mark survives. They also explain how detection is built into the Gemini app, how partners who share their keys can be detected through a central service, and how news organizations have used SynthID to check circulating images.
The second half turns to biology. Radcliffe describes SynthID Bio, a proof-of-concept that applies the same ideas to AI-designed biological sequences, so that sequences from AI systems can be told apart from those found in nature. The aim is to trace the origin of potentially harmful designs before anything is synthesized. The episode is a clear introduction to provenance, misinformation and biosecurity as generative AI expands across text, images, audio, video and molecules.
📺 Source: Google DeepMind · Published October 01, 2026
🏷️ Format: Podcast
