First AI-designed viruses a step towards AI-generated life (www.nature.com)

Researchers have used AI to design whole viral genomes for the first time, producing bacteriophages that can infect and kill Escherichia coli. Using two sequence models called Evo 1 and Evo 2—pretrained on over two million phage genomes and further fine‑tuned by supervised learning—the team generated thousands of ΦX174‑like single‑stranded DNA genomes (ΦX174 is a simple 5,386‑nt, 11‑gene virus used as a template). They synthesized 302 candidate genomes (most sharing >40% nucleotide identity with ΦX174, though some encoded novel protein sequences), recovered viable phages from host cells and found 16 designs with E. coli host specificity; combinations of AI‑designed phages could kill three E. coli strains that wild‑type ΦX174 could not. The work, posted as a bioRxiv preprint, demonstrates that generative models can produce coherent, genome‑scale sequences and points toward AI‑driven genome engineering and future “AI‑generated life.” Technically, it shows whole‑genome design is tractable with large sequence datasets and supervised fine‑tuning, with clear therapeutic implications for phage therapy against antibiotic‑resistant bacteria. At the same time the study raises biosafety and dual‑use concerns and requires further experimental validation and safeguards before clinical application.