AI Generates Fully Functional Bacterial Viruses, Opening New Therapeutic Frontiers

Stanford scientists employed generative AI systems called Evo1 and Evo2 to predict and assemble complete viral genomes, producing the first AI-designed viruses capable of replication. From an initial pool of 302 computational designs, 16 bacteriophages successfully killed E. coli bacteria in laboratory tests, a result that sparked spontaneous applause among the researchers. The AI was trained on genetic sequences from viruses, bacteria, plants and humans, but the team deliberately excluded any viruses that could infect complex organisms to mitigate risk.

Experts highlighted the dual-use dilemma, noting that while the method may enable new therapies against drug-resistant infections, it also poses serious biosecurity challenges. Commentators from Johns Hopkins University’s Center for Health Security emphasized the need for stringent safeguards to prevent malicious applications. The discovery signals a broader shift toward synthetic biology, where computer-driven genome design could eventually target a range of medical problems.