Google DeepMind unveils SynthID Bio to watermark AI-generated proteins

Google launched SynthID Bio on September 30, 2026, bringing digital watermarking to synthetic biology. The technology embeds imperceptible signatures directly into biological codes, allowing verification both in digital models and in synthesized physical proteins while preserving their biological function in lab tests.

In wet-lab tests targeting three proteins — VEGF-A, the SARS-CoV-2 spike protein RBD, and PD-L1 — watermarked designs performed comparably to unwatermarked versions in hit rate, binding affinity, and natural sequence diversity. Google said this created the first watermark-carrying protein binders that remain biologically functional.

For protein folding, SynthID Bio fine-tunes a small part of AlphaFold 3's diffusion network, building the watermarking capability directly into the model weights so predicted 3D coordinates carry detectable features. In collaboration with Stanford's Hie Lab and the Arc Institute, the system has also been integrated into the genomic model Evo 2 to watermark phage genomes designed by Evo 2, with early lab tests confirming the watermarked phages are functional.

Google describes SynthID Bio as a family of watermarking methods developed specifically for synthetic biology, aimed at strengthening biosafety and scientific integrity. The approach adjusts methods based on data type — subtly guiding amino acid choices for sequences and adjusting atomic coordinates for predicted 3D structures — to create reliable detection signals. The watermark layer is particularly relevant for DNA synthesis screening and can help maintain the integrity of databases such as the Protein Data Bank, UniProt, and GenBank. A key future challenge is making watermarks more robust against deliberate tampering.