Researchers at Stanford University and the Arc Institute have used generative artificial intelligence to design new viral genomes that functioned as bacteriophages in laboratory tests, according to the report published on August 24, 2026. The AI-designed viruses were able to infect and reduce Escherichia coli, or E. coli, bacteria. The work moves generative AI beyond producing text, images and software into designing biological systems.

The research used the Evo genome-language model series and employed bacteriophage ΦX174 as a biological reference. Researchers asked the AI to generate new genomes with functional structures rather than copies of a natural virus. About 300 designs were selected for synthesis and testing, and 16 became functional bacteriophages. Training data excluded genomes of viruses capable of infecting humans, while the designed viruses were intended to target bacteria rather than cause disease in people.

The findings remain a proof of concept because only a small share of proposed designs worked. The report identifies potential applications in phage therapy against antibiotic-resistant bacteria, but notes unresolved questions over effectiveness, safety, patient-specific phage selection and regulatory approval. Biosafety experts say security, oversight and controls must advance alongside genome-design capabilities.