Google DeepMind watermarks AI-designed proteins

Google DeepMind introduced SynthID Bio, a proof-of-concept watermarking system for AI-designed proteins. The idea is to put a detectable signature inside the biological sequence or predicted 3D structure itself, not just in a file or metadata layer.

DeepMind says the watermark can survive the jump from a digital design to a synthesized physical protein while preserving biological function. In lab tests on protein binders for VEGF-A, the SARS-CoV-2 spike RBD and PD-L1, watermarked designs matched unwatermarked designs on hit rate, binding affinity and sequence diversity. For protein folding, the team also fine-tuned part of AlphaFold 3 so predicted structures carry a detectable signal without hurting accuracy.

The practical reason this matters is biosecurity. AI-designed sequences may look unlike known organisms or known hazards, which makes normal DNA synthesis screening harder. A trusted-model watermark could give synthesis providers and databases another signal for provenance, helping them decide which designs need closer review.

DeepMind is publishing a methods paper, open-sourcing code and in vitro data, and releasing weights to researchers. The company says the next challenge is making the watermark more robust against deliberate tampering and extending it to more complex biological objects.