AI Designs Viruses That Kill Bacteria
Friday, 2026/08/07212 words3 minutes822 reads
Scientists have demonstrated that a genomic AI model can generate viable bacteriophage genomes: complete genetic blueprints for viruses that infect bacteria. The study used PhiX174, a compact phage whose biology has been investigated for decades, as a controlled test system.
The researchers fine-tuned genome language models, called Evo 1 and Evo 2, on phage sequences related to PhiX174. After the models generated candidate genomes, the team synthesized and tested selected designs. Sixteen of 285 tested candidates propagated in E. coli and inhibited growth of the target bacterial strains. The successful phages did not affect unrelated bacterial strains in those experiments.
This does not mean AI has designed a human virus, nor does it make phage therapy ready for routine clinical use. The work is an experimental demonstration with a relatively small bacterial virus. Still, it suggests that generative models may help researchers explore genetic designs more efficiently than trial and error alone.
Potential benefits include future tools against antibiotic-resistant infections. At the same time, the result sharpens a dual-use question: methods for designing useful biological systems can also create security concerns. The research team used a bacteriophage system partly because it could be tested with minimal biosecurity risk, but experts argue that governance, screening and responsible access must develop alongside the technology.
