AI Designs Novel Viruses: Medical Promise and Biosecurity Risks

On August 6, the journal Science published findings revealing that researchers have harnessed artificial intelligence to create viruses that do not exist in the natural world. The study, spearheaded by Brian Hie from Stanford University, employed generative AI models Evo 1 and Evo 2, which were trained on trillions of nucleotides. The focus was on bacteriophages, resulting in approximately 700,000 designs derived from the Phi X-174 phage, with 285 synthesized in the laboratory. Out of these, sixteen successfully inhibited the growth of E. coli, surpassing the effectiveness of natural phages. A mixture of these synthetic phages proved effective against antibiotic-resistant E. coli. While experts like Patrick Cai view this as a significant achievement, Isaac Bogoch raises concerns regarding biosecurity. Researchers took measures to exclude specific genomes from their training data and established biosafety protocols, yet worries about complexity persist.