PALO ALTO, Calif. — Scientists at Stanford University and the Arc Institute have used artificial intelligence to design and create functioning viruses from scratch, producing 16 bacteriophages capable of infecting and killing Escherichia coli in laboratory tests, a development that demonstrates the growing ability of AI to engineer complete biological genomes.
The viruses, which infect bacteria rather than humans, were generated using genome-language models known as Evo 1 and Evo 2. Researchers used the bacteriophage ΦX174 as a template and asked the models to generate complete viral genomes with characteristics needed for replication and infection. The study was led by Samuel King and Brian Hie of the Arc Institute and Stanford.
Researchers synthesized about 300 AI-generated genome candidates and tested them against E. coli. Sixteen produced viable bacteriophages, according to the researchers. Some performed better than the original ΦX174 in laboratory measures of bacterial killing, while a combination of the engineered phages was able to overcome resistance to ΦX174 in three E. coli strains.
The work could have implications for phage therapy, an approach that uses bacteriophages to target bacterial infections, particularly as bacteria become resistant to antibiotics and existing phages. The researchers said their results provide a framework for designing new bacteriophages against evolving bacterial pathogens.
The researchers also restricted the work to bacteriophages and excluded viruses known to infect humans, animals or plants from the relevant design process, according to reporting on the study. The resulting viruses therefore are not human pathogens, and the experiment was not designed to increase the ability of a human pathogen to cause disease.
The advance has nevertheless prompted biosecurity concerns because the same underlying capability could eventually be applied to more consequential biological systems. In an accompanying discussion, Johns Hopkins Center for Health Security researchers Tom Inglesby and Moritz Hanke warned that the ability to compose viral genomes with generative AI creates urgent biosafety and biosecurity questions.
Experts have called for safeguards extending beyond AI models to DNA synthesis screening, research oversight and laboratory containment. Researchers involved in the work have likewise acknowledged the need for biosafety and biosecurity measures as genome-scale AI design develops.
The study, first posted as a preprint in September 2025, remains a laboratory demonstration rather than an approved medical treatment. Further work would be required before AI-designed bacteriophages could be evaluated for therapeutic use in patients.
