Stanford and Arc Institute AI Models Design New Viruses

In a major breakthrough for synthetic biology, researchers from Stanford University and the Arc Institute have successfully used artificial intelligence to design completely new, functional viruses. Published in the journal Science, the study marks the first time a language model has generated fully working genomes from scratch. The research team utilized two models, Evo 1 and Evo 2, which were trained on millions of genomes to learn the fundamental language of biological life.

The AI models were tasked with generating novel variations of Phi X174, a well-documented bacteriophage that specifically targets and infects E. coli bacteria. Out of 285 synthesized and tested phages generated by the models, 16 proved to be viable. Some of these AI-designed viruses actually replicated at a faster rate than the naturally occurring original, and a few possessed genomes distinct enough to be classified as entirely new species.

To demonstrate the practical utility of these synthetic organisms, the researchers deployed a cocktail of the AI-generated viruses against E. coli bacteria that had developed resistance to the natural phage. The synthetic cocktail successfully eliminated the drug-resistant bacteria. To mitigate biosecurity risks, the scientists deliberately avoided training the models on any viruses that infect humans, animals, or plants, ensuring the generated outputs pose no threat to people.

For practitioners in biotechnology and medicine, this development signals a shift toward programmable therapeutics. Instead of hunting for rare natural phages to combat antibiotic-resistant superbugs, researchers can now use open-source models like Evo 2 to generate custom-tailored biological agents. However, because Evo 2 is publicly available, the breakthrough also intensifies the pressure on the scientific community to establish robust biosafety testing frameworks and guardrails to prevent the technology from being misused to engineer dangerous pathogens.