Google DeepMind Introduces SynthID Bio to Watermark AI-Designed Proteins

Google DeepMind introduced SynthID Bio on September 30, 2026, describing it as a proof of concept for watermarking methods in synthetic biology. The approach embeds an imperceptible signal in biological data, with the aim of making it verifiable in both digital models and synthesized physical proteins without compromising biological function.

The methods vary by data type: they subtly guide amino-acid choices in protein sequences and adjust atomic coordinates in predicted 3D structures. Google DeepMind tested protein binders designed with AlphaProteo alongside a SynthID Bio-enabled version of ProteinMPNN. In wet-lab tests involving VEGF-A, the SARS-CoV-2 spike protein RBD, and PD-L1, the company says watermarked designs matched unwatermarked versions in hit rate, binding affinity, and natural sequence diversity.

For protein folding, SynthID Bio fine-tunes part of AlphaFold 3’s diffusion network so predicted 3D coordinates carry a detectable signal. Google DeepMind reports that the method preserves prediction accuracy and offers near-perfect detectability, including after digital noise or minor coordinate changes. The work extends SynthID beyond an earlier application: Google DeepMind announced SynthID for watermarking AI-generated images in 2023.

Google DeepMind presents SynthID Bio as one verification layer for biosecurity and scientific integrity, not a standalone solution. The company identifies resistance to deliberate tampering and extending watermarking to more complex biological objects as continuing challenges and research areas.