AI-Designed Viruses Outperform Natural Phages Against Resistant Bacteria

Researchers at Stanford University and the Arc Institute have successfully used a genomic language model to design complete, functional viral genomes from scratch. This marks the first time AI has been used to generate entire functional viruses, moving the field beyond traditional genetic engineering toward true generative design.

Using the Evo 2 foundation model, the team generated 302 candidate genomes. The model was trained on more than 2 million bacteriophage genomes and further fine-tuned on 15,000 relatives of the target phage, ΦX174. Of the candidates produced, 16 functional bacteriophages—viruses that specifically infect bacteria—were successfully "rebooted" in a laboratory setting. Notably, these AI-designed phages were able to infect strains of E. coli that had evolved resistance to the natural viruses they were modeled after, a result that natural phage mixes failed to achieve.

The Evolution of Phage Therapy

Phage therapy has been employed for over a century as a method to treat bacterial infections. However, the practice has long been hindered by several systemic challenges, including narrow host ranges, the speed at which bacteria develop resistance, and complexities surrounding patenting. By utilizing generative AI, researchers can now explore "sequence space" more efficiently. Chase Beisel, cofounder of Locus Biosciences, noted that this approach provides a novel way to uncover new attributes that will be highly useful in the long run.

Despite the generative nature of the process, the resulting viruses remained closely tied to their origins. An independent analysis conducted by Oliver Crook of the University of Oxford revealed that the viable AI-generated genomes were, on average, 97% identical to the ΦX174 template.

Implications for Medicine and Security

This breakthrough offers a potential path toward bespoke, AI-generated phage therapies designed to target antibiotic-resistant "superbugs." By tailoring viruses to specific bacterial mutations, medicine could overcome the rapid resistance cycles that render current antibiotics ineffective.

However, the ability to design functional genomes also introduces significant biosecurity risks. Experts warn that the same tools used to create life-saving medicines could lower the technical barrier for the creation of dangerous biological agents. The capacity to synthesize functional viruses from digital designs necessitates a new approach to biological oversight.

The Path Forward

As the technology matures, the focus is shifting from technical feasibility to ethical governance. Biosecurity experts at the Johns Hopkins Center for Health Security stated that the primary question is no longer whether generative viral genome design will exist, but whether society can establish oversight frameworks that allow medical benefits to unfold while preventing the technology from enabling serious harm.