Researchers from Stanford University and the Broad Institute of MIT and Harvard demonstrated that an AI model named Evo can design entirely new viral genomes from scratch.
The team trained Evo on large-scale genomic data, enabling it to learn the underlying rules of viral sequence architecture. They then generated thousands of candidate genome sequences and synthesized nearly 300 designs for laboratory testing. Of these, 16 produced fully functional bacteriophages—viruses that infect E. coli bacteria and are harmless to humans.
This is the first recorded instance of AI designing viable, self-replicating viruses that do not exist in the natural world. While the phages themselves pose no direct threat, the study's authors explicitly warned that the methodology could be misapplied to design pathogenic viruses. The paper calls for "urgent international dialogue" on AI biosecurity protocols, noting that current regulatory frameworks lag far behind the rapid pace of generative AI development.
The researchers also released a responsible-use guideline for AI-driven synthetic biology, urging that model weights and training data be kept under restricted access.
The breakthrough has already prompted calls from biosecurity experts for mandatory screening of AI-generated nucleic acid sequences before synthesis.