Stanford’s Evo 2 Designs E. Coli-Killing Phages, Stoking Biosecurity Fears

Bacteriophages are viruses that infect and kill bacteria, making them a potential alternative to antibiotics as drug resistance spreads. Stanford University and the Arc Institute developed Evo 2, a genome language model that can generate complete DNA sequences much as a text model predicts words. The technology could accelerate the design of phage therapies tailored to specific pathogens, but its ability to write functional viral genomes also raises concerns that similar systems could eventually be misused to engineer dangerous pathogens.

In research published in Science on Aug. 6, 2026, the team used the E. coli-infecting phage ΦX174 as a starting point and synthesized and tested 302 AI-designed genomes. Sixteen produced viable phages, with some outperforming natural ΦX174, while a cocktail of all 16 suppressed E. coli strains resistant to the native virus. The researchers excluded viruses that infect humans, animals and plants from relevant training data, but biosecurity specialists said oversight and DNA-synthesis screening have not kept pace with generative genome technology.