AI Successfully Designs Entirely New, Functional Viruses in Historic First
American researchers have successfully used generative artificial intelligence to design 16 functional, laboratory-replicable bacteriophage viruses that destroy E. coli bacteria. While celebrated as a historic breakthrough in synthetic biology and medicine, experts warn that AI-designed biological entities raise urgent biosecurity concerns.
American researchers have used artificial intelligence to design fully functional, lab-replicable viruses for the first time, marking a milestone in synthetic biology, according to the BBC. It's the first time an entire genome has been generated using generative AI — producing 16 novel viruses capable of infecting bacteria, which researchers confirmed pose no threat to humans.
Scientists see the achievement as a potential turning point for medical treatment, but experts warn AI-designed biological agents raise urgent biosecurity and safety concerns. AI is already being used to help discover new antibiotics, but building a viable, entirely new virus from scratch is a far more complex undertaking.
"This is the next step in what can be designed using generative AI; it is the first time generative AI has been used to design a full genome," Stanford Assistant Professor Brian Hie told the BBC. He noted the resulting entities can replicate inside host cells and carry out other biological functions — opening a new realm of experimentation for the team.
The underlying approach works similarly to large language models like ChatGPT, which predict sequences of text. But the models used here, Evo 1 and Evo 2, predict what researchers call the "language of life" rather than human words. Trained on genomic data from viruses, bacteria, plants, and humans, the models were designed specifically to create bacteriophages — viruses that target and kill specific bacterial strains.
From 302 promising AI-generated designs, Stanford researchers synthesized viruses in the lab. Sixteen successfully infected and killed E. coli bacteria. Since viruses aren't considered fully living organisms, researchers stress that creating actual living cells with AI would involve far greater evolutionary hurdles.
A phage's genetic code spans roughly 5,400 base pairs, compared to about 500,000 in the smallest known living-cell genome and three billion in the human genome. Hie said attempting to create simple living cells this way would be a massive undertaking — but one he believes remains possible, and one the team intends to pursue.
Professor Marc Güell of the Synthetic Biology Laboratory at Pompeu Fabra University in Spain called the study a historic milestone, noting that for the first time, humanity has begun designing biological entities directly through computational systems.