STANFORD, California: AI-designed viruses killed resistant E.coli strains in laboratory tests, researchers reported Thursday, August 6, in Science.
The peer-reviewed study marked the first generative design of complete bacteriophage genomes, according to its authors. Bacteriophages infect bacteria and cannot directly infect human, animal or plant cells.
Samuel H. King, a Stanford bioengineering doctoral student, led the work with chemical engineering professor Brian L. Hie.
Meanwhile, researchers used Evo 1 and Evo 2 with ΦX174, a small bacteriophage, as their design template. The models generated thousands of genome sequences targeting nonpathogenic E. coli C bacteria.
Researchers chose 302 candidates for synthesis and successfully assembled 285 for laboratory testing. Ultimately, 16 designs produced viable phages with different genetic sequences, structures and laboratory fitness profiles.
Some phage combinations overcame resistance in two E. coli strains that resisted ΦX174-like viruses, Science reported.
Read: Shrinking Human Attention Span: How Digital Distraction Is Rewiring the Brain
However, the result remains laboratory proof of concept and does not establish a human treatment. AI produced genetic sequences, while researchers synthesized and tested the physical genomes.
The researchers excluded viruses infecting humans, animals, plants and fungi from the models’ specialised training data. They also urged future whole-genome projects to involve safety and security experts throughout development.
Separately, a companion Science commentary described functional AI-generated viral genomes as an urgent biosafety and biosecurity issue.
Read: Neuralink to Begin Mass Production of Brain Implants in 2026: Elon Musk
Authors Thomas V. Inglesby and Moritz S. Hanke work at the Johns Hopkins Center for Health Security. They warned against designing viruses capable of causing disease in humans, animals or plants.
Meanwhile, the research team said model safeguards and existing safety frameworks could reduce misuse risks.
