AI Designs 16 New Viruses From Scratch to Fight Drug-Resistant Bacteria

Stanford researchers used AI models Evo 1 and Evo 2 to design functional bacteriophages from scratch — a breakthrough for phage therapy that also raises pressing biosecurity questions.

Researchers from Stanford University and the Arc Institute have crossed a striking threshold in generative biology: they used AI to design entirely new viral genomes from scratch — and 16 of them turned out to be functional, self-replicating viruses.

The AI models, Evo 1 and Evo 2, were trained on DNA sequences the way large language models learn from text. The team focused on a bacteriophage called ΦX174 — a virus that infects E. coli and cannot harm humans. After fine-tuning the models on thousands of related viral genomes, they generated hundreds of candidate sequences. Out of 285 designs tested in the lab, most failed. But 16 survived and could infect bacteria, reproduce, and produce new viral particles.

Some of the AI-designed phages performed better than the natural ΦX174 in competition experiments. A cocktail of the synthetic viruses managed to overcome bacterial resistance that had defeated natural phages entirely — a result that points toward phage therapy, a century-old approach gaining renewed urgency as antibiotic-resistant infections claim more lives each year.

The study, published in the journal Science, also carries an uneasy footnote. Evo 2 is fully open-source. Anyone can download it. Turning a computer-generated sequence into a functioning organism still requires lab equipment and genuine expertise — but that distance is shrinking. Biosecurity researchers have warned that the ability to design viral genomes is advancing faster than the governance meant to control it. The concern is not that this experiment created something dangerous — it's that increasingly capable biological AI will lower the barrier for others who might try.

The 16 phages in this study are harmless to humans. But the proof of concept is now public, and unlike an AI-generated image, a successful genome can become something that reproduces.

Sources: Ars Technica, ZME Science, The Guardian