Introducing SynthID Bio

SynthID Bio introduces a method to embed imperceptible watermarks directly into the biological code of AI-generated proteins, ensuring the signature remains detectable even after physical synthesis. The technology adapts to different data types by subtly guiding amino acid choices and adjusting atomic coordinates in predicted 3D structures, creating a reliable detection signal. Experiments confirmed that these adjustments did not compromise the protein’s biological function, a critical requirement for therapeutic and research applications.

Researchers tested SynthID Bio on protein binders designed to target specific proteins, using AlphaProteo and a SynthID Bio-enabled version of ProteinMPNN. In wet-lab experiments across three targets—VEGF-A, the SARS-CoV-2 spike protein RBD, and PD-L1—the watermarked designs matched the performance of unwatermarked versions in hit rate, binding affinity, and sequence diversity. This marks the first successful creation of biologically functional, watermarked protein binders.

For protein folding predictions, SynthID Bio fine-tunes AlphaFold 3’s diffusion network to embed watermarks directly into the model’s weights, ensuring inherent detectability in predicted 3D coordinates. The approach maintains AlphaFold 3’s prediction accuracy while achieving near-perfect detectability and preserving key structural features, even under digital noise or minor coordinate changes.

SynthID Bio is positioned as a verification layer within a broader biosecurity framework, complementing existing safeguards like model-level mitigations and customer vetting. It aims to assist DNA synthesis providers in screening AI-generated sequences, reduce manual reviews, and help maintain the integrity of public biological databases such as the Protein Data Bank and UniProt.