AI Designs 16 New Viruses That Kill Resistant Bacteria
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- Stanford University and the Arc Institute used AI models Evo 1 and Evo 2 — trained on millions of genomes from all domains of life — to design 16 previously unknown bacteriophages capable of infecting and eliminating bacteria.
- The researchers used the Phi X-174 bacteriophage, which infects E. coli, as a reference architecture; of 300 AI-designed synthetic genomes synthesized in the lab, only 16 produced fully functional viruses featuring previously unpublished sequences, different genes, new regulatory elements, and varying genome sizes.
- The AI-generated phages overcame bacterial resistance in E. coli strains faster than natural phages similar to Phi X-174, demonstrating what the authors call a path toward AI-designed phage therapies against rapidly evolving pathogens.
- The study was published this week in the journal Science and represents the first time an AI has created functional viruses from scratch rather than replicating known pathogens or variants.
- Moritz Hanke of the Johns Hopkins Center for Health Security warned there are currently no safeguards capable of effectively preventing AI-assisted creation of a lethal virus, citing a "huge disconnect" between the speed of scientific advancement and regulatory development.
- The Rand Corporation had warned three years ago that advanced AI systems could refine the planning and execution of biological-weapons attacks, with fears now growing that such capabilities will become more sophisticated.
Why it matters: Stanford and Arc Institute's demonstration gives drug developers a new tool against the growing crisis of bacterial resistance, with the AI-designed phages infecting resistant E. coli faster than natural alternatives. But the same capability — generative AI designing functional biological entities from scratch — outpaces existing safeguards, according to Johns Hopkins' Hanke, who told The New York Times no effective oversight currently exists to prevent malicious use.
