SynthID Bio
Google DeepMind introduced research methods for marking AI-generated protein sequences and predicted structures so their origin can be checked.
What happened lately
SynthID Bio timeline
A mark in a biological design
Google DeepMind introduced SynthID Bio on September 30 as a proof of concept for tracing AI-generated biological designs. The company describes related methods for protein sequences and predicted three-dimensional structures. For sequences, it changes the choice of amino acids so that a signature can be detected even after a protein is synthesized. For structures, it adjusts predicted coordinates through a modified AlphaFold 3 model.
Evidence and limits
DeepMind reports laboratory tests of watermarked protein binders against three targets and says their measured function matched unwatermarked designs. Those are the company’s research results, not an independent guarantee that every protein design can be marked without changing its properties. DeepMind says resistance to deliberate watermark removal remains a challenge. A detectable origin signal may help screening and database labeling, but it does not by itself show that a biological design is safe.