Stanford AI Designs First Complete Viral Genomes
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- Stanford University researchers designed 16 working bacteriophages using AI models Evo1 and Evo2, with 16 of 302 synthesized designs proving effective at killing E. coli — the first complete viral genomes made by generative AI.
- Brian Hie, Stanford assistant professor, said the work was "new territory" because generative AI had previously only designed simpler molecules like antibiotics, never genomes that can replicate and function inside cells.
- Samuel King, the PhD student who ran the lab tests, said the team spotted the phages working in the early hours of the morning and the room "spontaneously burst into applause" when results were shared with the wider team.
- Johns Hopkins Center for Health Security authors Dr. Thomas Inglesby and Dr. Moritz Hanke warned in a Science commentary that the findings raise "urgent biosafety and biosecurity questions" and said viruses that could cause disease "should not be pursued."
- Stanford researchers built in safeguards: they excluded viruses that infect complex organisms from the training database, tested only phages rather than human-infecting viruses, and conducted the work in a secure laboratory.
- Prof Marc Güell of Pompeu Fabra University called the study a "very significant turning point" because "for the first time in history, we are beginning to design biology on a computer."
- Prof Patrick Cai of the Manchester Institute of Biotechnology called it an "important milestone" suggesting genome language models are "beginning to learn the design principles encoded by evolution."
Why it matters: The breakthrough offers a potential new weapon against antibiotic-resistant infections — a growing global health crisis — by enabling custom phage therapies on demand. But because the same technique can theoretically be extended to viruses that infect humans, the Johns Hopkins authors warn the field now has weeks-to-months to set guardrails before the technology reaches bad actors.
