First AI-Designed Viruses Kill Drug-Resistant E. Coli
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- Dr. Brian Hie at Stanford University used genome language models Evo1 and Evo2, trained on 2 million bacteriophages, to design functioning viral genomes — a first reported in the journal Science.
- Researchers selected roughly 300 AI-generated genomes to synthesize in the lab; only 16 proved viable, but a cocktail of them overcame resistance in two strains of E. coli that natural bacteriophages could not kill.
- The training data intentionally excluded genetic code from viruses that infect plants, humans, or other animals to reduce the risk of the AI producing dangerous pathogens.
- Prof. Tom Inglesby and Dr. Moritz Hanke at Johns Hopkins's Center for Health Security wrote in an accompanying commentary that "the ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not."
- Prof. Tom Ellis of Imperial College London called the work impressive but said the threat of full AI genome design is "very overblown" compared to existing gain-of-function modifications of known pathogens.
- Dr. Filippa Lentzos of King's College London argued governance should target the DNA-manufacturing stage, not just the AI model itself, calling for a "layered approach" spanning model safeguards, research review, and synthesis screening.
Why it matters: The paper demonstrates that generative AI can now produce functioning viral genomes from scratch — a capability biosecurity experts say outpaces existing governance, with Inglesby and Hanke explicitly warning such work on human-infecting pathogens should not be pursued. For phage therapy, the cocktail's success against resistant E. coli is a clinical proof-of-concept, but only 16 of roughly 300 designed genomes proved viable, underscoring how far the technology sits from routine therapeutic use.
