Particle.news

AI Designs and Lab‑Makes 16 New Bacteriophages

A Science paper shows generative genome models can turn digital sequences into working viruses and has prompted calls for tighter biosecurity and oversight.

Overview

  • The peer‑reviewed study published Friday reports that researchers at Stanford and the Arc Institute used genome‑language models called Evo 1 and Evo 2 to generate about 700,000 candidate viral genomes and that 16 synthetic bacteriophages proved viable in lab tests.
  • The team narrowed designs to roughly 285–302 top candidates, chemically synthesized those sequences, inserted them into E. coli and observed 16 new phages that replicated and killed bacterial cells.
  • Researchers say they limited training data to exclude viruses that infect animals or plants, ran experiments in high‑containment labs, and that Evo 2 has been released as open‑source, a move that has intensified debate over access controls.
  • Public health experts including authors of a Johns Hopkins commentary warned the result raises urgent biosecurity questions and urged legal limits, required safety review for generative‑genomics work, and faster governance development.
  • The result points to possible tailored phage therapies against drug‑resistant bacteria but also shows clear limits today: the genomes were very small, only a small fraction of AI designs were viable, and the models lack mechanistic explainability, so further technical work and policy safeguards are needed.