Google DeepMind Brings Digital Watermarking to Synthetic Biology With SynthID Bio

September 30 (IT Home) — Google today introduced SynthID Bio, bringing digital watermarking technology to the field of synthetic biology. The technology can embed an imperceptible signature directly into biological code, and it can be verified not only on digital models but also on synthesised physical proteins, while keeping the proteins’ biological function in laboratory tests.

Generative AI is helping scientists solve key biological challenges, from predicting protein structures to designing entirely new proteins and even developing new bacteriophages. But these tools also bring new challenges: newly AI-designed sequences may bypass traditional DNA synthesis screening, and mislabelled synthetic 3D structures can contaminate public databases and mislead later research.

SynthID Bio is a family of watermarking methods developed specifically for synthetic biology, intended to strengthen biosecurity and scientific integrity.

It adapts its method to the type of data, subtly guiding the choice of amino acids in a sequence or adjusting the atomic coordinates of a predicted 3D structure, so as to create a reliable detection signal. Experiments show that these adjustments do not harm the proteins’ biological function, which is essential for effectively treating disease and advancing scientific research.

In wet-lab tests against three target proteins — VEGF-A, the SARS-CoV-2 spike protein RBD and PD-L1 — the watermarked designs matched unwatermarked versions in hit rate, binding affinity and natural sequence diversity, producing the first watermarked protein binders that retain biological function.

For protein folding, SynthID Bio fine-tunes a small part of AlphaFold 3’s diffusion network, building watermarking capability directly into the model weights. This ensures that the predicted 3D coordinates themselves carry a detectable signature. No matter who runs the model, SynthID Bio preserves AlphaFold 3’s prediction accuracy while providing near-perfect detectability, maintaining the distribution of key structural features and withstanding digital noise or small coordinate changes.

Biosecurity relies on layered defences, and SynthID Bio’s watermarking method can serve as an important verification layer embedded in the biological design itself. This layer is especially important for DNA synthesis screening: AI can now create entirely new sequences that bear almost no resemblance to known hazards, so screening staff can no longer rely on old assumptions — and SynthID Bio can provide an automatic verification signal showing that an order came from a trusted model with built-in safeguards.

Similarly, SynthID Bio can help maintain the integrity of databases such as the Protein Data Bank, UniProt and GenBank, ensuring that synthetic entries are correctly flagged, or marked for further review.

Google says that although no single biosecurity intervention is a panacea, SynthID Bio brings the proven watermarking tool SynthID into synthetic biology, and is an important first step towards reliably identifying and tracking AI-generated biological sequences and structures.

Key challenges ahead include making the watermark more robust against deliberate tampering. SynthID Bio could also be used alongside provenance metadata methods or a central repository of AI-generated biological data to better identify and track AI-generated proteins.

Google is currently studying how to apply watermarking to more complex biological objects. In research with Stanford’s Hie lab and the Arc Institute, it has already integrated SynthID Bio into the advanced genomic model Evo 2, watermarking phage genomes designed by Evo 2, and early laboratory tests have confirmed that these watermarked phages are functional.