Technology & AINews date: Published by Svertok:
Google Introduces a Watermark for AI-Designed Proteins
SynthID Bio places a verifiable signal in generated protein sequences and predicted structures while seeking to preserve biological function.

Google's SynthID Bio page was published on September 30, 2026. The company introduced the technology as a way to watermark AI-designed proteins while preserving their biological function.
SynthID Bio embeds an imperceptible, verifiable signal directly in biological designs, including generated protein sequences and predicted three-dimensional structures. That signal is meant to let researchers check a design's provenance without relying solely on a separate label or database record.
What the laboratory tests found
Google reports that its watermarked designs matched the performance and natural diversity of versions without the watermark in laboratory tests across target proteins. A binding-affinity comparison covered three targets and used KD, for which a lower measurement indicates stronger binding.
The examples include a watermarked VEGF-A protein binder and a 7PPA comparison showing an AlphaFold 3 predicted structure, the ground truth structure and the watermarked structure. These results cover the tested proteins and structures; Google does not establish equivalent performance for every protein, biological task or deployment setting.
Google says the watermark could add a provenance layer that strengthens biosecurity and protects the integrity of open scientific databases. The announcement does not provide an adoption plan, independent validation results or a public release schedule.
Next

Google Opens SynthID Detector to the Public for AI-Media Checks
The English-language tool can look for SynthID signals in images, video and audio from Google and participating partners, a scope that does not cover every AI-made file.
Read next
