AI-Designed Viruses Kill Drug-Resistant Bacteria in Lab First

First functional viruses designed by AI

Researchers at Stanford University have successfully created the first functioning viruses designed entirely by artificial intelligence, according to findings published in the journal Science. Dr. Brian Hie, a chemical engineer, used genome language models—genetic equivalents of the large language models powering chatbots—to design complete genomes for bacteriophages, viruses that exclusively target bacteria.

In laboratory testing, a combination of these AI-designed bacteriophages killed strains of E. coli that had developed resistance to naturally occurring phages. The breakthrough demonstrates AI's potential to rapidly design and customize viral genomes to combat specific bacterial threats.

Why it matters

This development arrives at a critical moment as antibiotic resistance threatens to make common infections untreatable. The ability to quickly design bacteriophages tailored to resistant bacteria could provide a powerful new tool against superbugs. However, the same technology that enables therapeutic advances also creates pathways for designing harmful pathogens—a dual-use dilemma that currently lacks adequate governance frameworks.

How the technology works

Hie's team employed AI models called Evo1 and Evo2, trained on genetic data from 2 million bacteriophages. Crucially, the researchers excluded genetic code from viruses capable of infecting plants, humans, or other animals to minimize risks of creating dangerous pathogens.

The AI generated thousands of candidate genomes. Researchers selected nearly 300 for laboratory synthesis, inserting them into bacteria that then produced the new bacteriophages. The success rate proved modest—only 16 viable viruses emerged—but a cocktail of these successfully overcame resistance in two E. coli strains.

Urgent biosecurity warnings

Alongside the scientific achievement, both the Stanford team and independent experts issued stark warnings about biosecurity implications. In an accompanying article, Professor Tom Inglesby and Dr. Moritz Hanke from Johns Hopkins University's Center for Health Security wrote: "The ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not."

The researchers emphasized that while bacteriophage genomes are tiny, the work proves generative AI can create functioning viral genomes. They explicitly cautioned against pursuing similar work with pathogens capable of infecting humans, animals, or plants, noting such genomes "might encode new pathogens that cannot be contained by existing countermeasures."

Expert perspectives on risk

Tom Ellis, a professor of synthetic genome engineering at Imperial College London, acknowledged the work's significance while noting bacteriophages represent "literally the smallest and easiest genome to make." He argued that the threat from full AI genome design is "very overblown" compared to the more accessible danger of modifying existing pathogens through gain-of-function research.

Dr. Filippa Lentzos from King's College London emphasized the need for layered governance rather than focusing solely on AI models. She identified DNA synthesis as a critical intervention point, calling for "safeguards around model development and access, responsible research review, synthesis screening, and established laboratory biosafety and biosecurity."

The details were first reported by The Guardian.