AI Designs Functional Viruses

Researchers at Stanford University and the Arc Institute, in collaboration with the Broad Institute, used generative AI models Evo1 and Evo2 to design complete viral genomes for the first time. The AI, trained on genetic sequences from millions of organisms, generated thousands of potential genomes for bacteriophages similar to Phi X 174. Of 285 synthesized designs, 16 produced viable viruses that could infect and kill E. coli bacteria, with some multiplying faster than natural phages. The study, published in Science, marks a significant milestone in synthetic biology, demonstrating that AI can design functional biological systems. While the viruses are harmless to humans, experts including those at the Johns Hopkins Center for Health Security raised urgent biosafety and biosecurity concerns, noting that governance has not kept pace with the technology. The researchers took precautions by excluding human-infecting viruses from training data and conducting work in secure labs. The breakthrough could lead to new phage therapies for antibiotic-resistant infections, but also raises the possibility of AI being misused to create dangerous pathogens.