Google DeepMind's SynthID watermarking extends to DNA sequences

Google DeepMind has expanded SynthID, its watermarking technology, beyond text, images and video into biological sequences. The system embeds invisible mathematical watermarks that allow creators to prove AI provenance and detect misuse without compromising the underlying function of generated proteins or other biological data.

In a new video, VP of Science Pushmeet Kohli and Research Scientist Jeremy Ratcliffe explain how watermarking works across different media types and why it matters for biosecurity. The approach addresses a growing concern: as AI becomes better at generating biological sequences, distinguishing between human-designed and AI-generated biology becomes harder—creating potential risks for dual-use research.