AI designs first complete viruses, raising hope and biosecurity fears

Scientists at Stanford University and the Arc Institute in California used artificial intelligence to design complete bacteriophage genomes from scratch, a result that pushes virus design ahead of the biosafety rules meant to contain it. The project generated thousands of candidate genomes, about 300 were built and tested in the lab, and 16 proved viable, marking the first time whole genomes have been successfully designed by AI.

The viruses were bacteriophages, so they infect bacteria rather than people, and they were built to target specific bacterial hosts. That makes the work potentially useful for phage therapy and for research on antibiotic-resistant bacteria, where new ways to direct viruses at hard-to-treat microbes could open a faster path to treatment design.

The same advance has intensified biosecurity worries. Experts reacting to the work said it raises urgent biosafety and security questions, and that catastrophic misuse cannot be ruled out if powerful models and lab access fall into the wrong hands. The concern is not the bacteriophages themselves, but the growing ability of AI systems to generate biological designs that can be translated into the real world.

Arc Institute background material says its Evo series of genome foundation models was built to move from designing individual genes to designing complete genomes, a sign of how quickly the field has shifted from analyzing biology to generating it. That shift matters because software-generated designs can, in principle, become physical organisms once they reach a laboratory.

The result leaves governments, universities and research institutions facing a sharper governance problem than the science headline alone suggests. The question now is where legitimate medical research ends and dual-use capability begins, and whether screening systems, lab oversight and access controls can keep pace with models that can already draft complete viruses.