08/11/2026
AI Has Designed More Than a Dozen New Viruses That Infect Bacteria
Scientists have used an artificial intelligence model called Evo to create more than a dozen previously unknown bacteriophages, viruses that infect bacteria.
Evo is a biological language model trained to recognize patterns in DNA. Similar to how a language model predicts the next word in a sentence, Evo predicts which genetic letters, A, C, G, or T, should come next in a DNA sequence.
Researchers trained the system on thousands of bacteriophage genomes and instructed it to generate new versions of ΦX174, a small virus that infects E. coli.
The team then chemically synthesized 285 AI generated genomes and tested them. Sixteen successfully developed into functional viruses capable of infecting E. coli, reproducing inside the bacteria, and killing the infected cells.
These viruses were not simply copies of naturally occurring viruses. Their genomes contained between 67 and 392 mutations compared with their closest known relatives. Thirteen also contained genetic changes that had never been observed in known natural sequences. One was sufficiently different that it could potentially represent a new viral species.
The researchers chose ΦX174 because of its place in genetic science. In 1977, it became the first organism to have its entire genome sequenced. In 2003, it became the first genome assembled completely from synthetic DNA.
Now, the same virus has served as a template for functional genomes generated with the help of AI.
The newly created viruses were also highly selective. They infected two closely related laboratory strains of E. coli but failed to reproduce in six other bacterial strains tested. Because they are bacteriophages, these viruses do not infect human cells.
Researchers also combined several of the AI designed viruses and tested them against three E. coli strains that had developed resistance to the natural ΦX174 virus. The AI generated combinations successfully overcame all three resistant strains, while the original virus did not.
Although this technology could eventually contribute to new approaches for treating antibiotic resistant bacterial infections, it also raises major biosecurity concerns.
Scientists warn that as biological AI systems become more powerful, similar technologies could potentially be misused to design viruses or other biological agents capable of affecting humans, animals, or crops.
Study: “Generative design of bacteriophages with genome language models,” Science