Introducing SynthID Bio

Google DeepMind says it has built SynthID Bio, a watermarking system for synthetic biology that hides an imperceptible signature inside AI-generated proteins and related structures. The company says the mark survives the jump from model output to a real, synthesized protein while leaving biological function intact in lab tests.

That matters because AI protein design is getting powerful fast. DeepMind points to its own work on protein structure prediction and new protein design, alongside newer efforts like bacteriophages. The same tools that help researchers build useful molecules can also produce novel sequences that don’t match anything in existing screening databases, which makes old safety checks less reliable.

The system takes different routes depending on what’s being watermarked. For protein sequences, it nudges amino-acid choices. For predicted 3D structures, it adjusts atomic coordinates. In experiments on protein binders, DeepMind says the watermarked versions matched the unwatermarked ones on hit rate, binding affinity and natural sequence diversity across three targets: VEGF-A, the SARS-CoV-2 spike protein RBD and PD-L1.

The company also says it can watermark folding outputs by fine-tuning part of AlphaFold 3’s diffusion network, so the signature is built into the model weights themselves. DeepMind says this preserves prediction accuracy, stays detectable even with digital noise or small coordinate changes, and holds up well enough to be useful as a provenance signal.

The biosecurity pitch is blunt: watermarking could help DNA synthesis providers, database curators and reviewers tell trusted AI-made designs from everything else. DeepMind is publishing the methods paper, open-sourcing code and in vitro data, and releasing weights to the research community. It is also saying the next step is harder: making the watermark harder to tamper with, and extending the idea to more complex biological objects, including an AI-designed bacteriophage in work with Stanford’s Hie lab and Arc Institute.