Researchers Use AI to Design Functional Synthetic Viruses for the First Time

US researchers from Stanford University, the Broad Institute of MIT and Harvard, and the Arc Institute have used artificial intelligence to design entirely new, functional viruses. The findings, published in the journal Science, mark the first time generative AI has been used to design complete genomes that can replicate and function inside cells. The team used AI models known as Evo1 and Evo2, which were trained on genetic codes from plants, people, bacteria, and viruses, to predict the sequences for bacteriophages. These are a specific type of virus that infects only bacteria and poses no threat to humans.

Out of 302 AI designs that were synthesized in the laboratory, 16 proved effective at killing E. coli bacteria. In some tests, a mixture of these synthetic viruses was more effective at killing the bacteria than naturally occurring viruses. The researchers noted that this breakthrough could lead to new treatments for antibiotic resistant infections, as phages can attack superbugs that are immune to traditional drugs.

Despite the medical potential, the study has sparked urgent warnings regarding biosafety and biosecurity. Experts from the Johns Hopkins University Center for Health Security stated that while the application is promising, there is currently no established governance to safely control the ability of generative AI to compose viral genomes. Some researchers emphasized that the risk is not the specific viruses created in this study, but the fact that AI can now perform this task at all.

Outlets with a center lean focused on the scientific process and the distinction between simple antibiotics and complex genome design. Outlets with a center left lean placed more emphasis on the potential for bioterrorism and the need for government oversight, with one report linking the event to recent reports of autonomous malicious activity by frontier AI models.