AI
AI Designs 16 Entirely New Viruses That Infect Bacteria, Raising Security Alarms
Researchers used machine learning to generate novel bacteriophages from scratch, opening doors to phage therapy but also amplifying biosecurity concerns.
WIRED — AI
Scientists at Stanford University and the Arc Institute have achieved a first: deploying artificial intelligence to generate 16 previously unknown viruses capable of infecting and killing specific bacteria. While the work points toward novel treatments for antibiotic-resistant infections, it simultaneously underscores growing fears about AI's potential to enable the creation of dangerous biological agents.
How the Breakthrough Works#
Researchers used two foundational AI models called Evo 1 and Evo 2—systems trained on millions of genomes across all life domains—to understand evolutionary patterns in viral and genetic organization. Rather than merely copying known pathogens, the algorithms learned how genes are typically arranged, which sequences nature preserves, and the biological rules that keep organisms functional.
The team focused on bacteriophages, viruses that exclusively target bacteria and are far simpler to work with than pathogens infecting humans. They used the bacteriophage Phi X-174—which naturally infects E. coli—as a structural template only, instructing the AI to create thousands of entirely novel genomes capable of recognizing and infiltrating the same bacterial host.
The resulting genomes retained the functional architecture needed to recognize bacteria, insert DNA, replicate it, and assemble new viral particles—but the specific DNA sequences diverged sharply from anything found in nature. The researchers synthesized 300 candidate genomes in the lab and tested them against E. coli. Sixteen proved fully functional, each displaying distinct genetic features, different regulatory elements, and varying sizes.
Outperforming Natural Viruses Against Resistant Bacteria#
The team then challenged these AI-designed phages against E. coli strains that had already developed resistance to the original Phi X-174 virus. The artificial variants rapidly overcame that resistance and established infection—a result the researchers published in Science as evidence that AI-generated therapies could evolve alongside rapidly mutating pathogens.
This capability could enable personalized, adaptive treatments that match the speed at which bacteria develop defenses, potentially transforming how clinicians combat resistance in real time.
