Stanford scientists used AI to design brand-new virus genomes, and 16 already kill E. coli in lab tests
Stanford researchers used the generative AI model Evo 2 to write entirely new bacteriophage genomes from scratch. Sixteen of nearly 300 AI-designed phages proved able to kill E. coli in lab tests, and combined into a cocktail they overcame bacteria resistant to the natural virus.
Stanford researchers had the generative AI model Evo 2 write entirely new bacteriophage genomes from scratch, starting from a snippet of the natural virus phi X 174. Nearly 300 AI-designed phage genomes were synthesized and tested against E. coli; 16 worked well enough to kill the bacteria, some replicating faster than the natural template. Combined into a cocktail, those 16 phages overcame E. coli strains already resistant to the original virus, a proof of concept for a resistance-resistant antibiotic. The work, led by chemical engineer Brian Hie and PhD student Samuel King, was published in the journal Science, and Evo 2 was released open source. Researchers covering the release have also raised biosecurity questions, since no public rulebook yet governs AI tools that can compose entire viral genomes, even though the system currently stays capped at small, phage-sized genomes far short of what would be needed to design a human-infecting virus. Sources: Stanford Report (news.stanford.edu, Aug 6 2026), CNN, Inside Precision Medicine.