A Stanford-led research team has used generative AI to design synthetic viruses — the first time artificial intelligence has been used to create an organism never before seen in nature — according to a study published in Science4.
The AI-designed viruses are bacteriophages, a class of virus that only infects bacteria and is already used worldwide to treat patients with persistent infections6. In lab tests, a cocktail of the AI-designed viruses killed E. coli bacteria that were resistant to natural bacteriophages. The research was conducted by researchers based at Stanford University5.
The work builds on large genome models originally developed to process DNA sequences. These models had previously demonstrated the ability to output DNA sequences that could encode functional proteins in bacteria and mimic gene structures found in complex cells. The same models were then used to output complete genomes of viruses that infect bacteria. All the viruses the models created are closely related to an existing virus, but they possess distinct features that would be challenging to produce through natural evolution.
The breakthrough carries dual implications. The technology opens new possibilities for combating bacterial resistance2. But it also raises concerns about the pace at which the capability is outstripping regulation. The researchers themselves suggest that the scientific community should begin preparing now for the possibility that someone could develop a related AI capable of designing a virus that targets vertebrates.
The regulatory landscape is already in flux. The Trump administration last month issued a policy for stopping high-risk research in the life sciences. That policy prohibits federally funded "gain of function" research and calls for enhanced oversight of projects involving harmful biological agents. However, the AI-generated viruses represent something categorically different from gain-of-function work on natural pathogens — the AI created something entirely new rather than modifying an existing organism.
Johns Hopkins health security experts Thomas Inglesby and Moritz Hanke wrote in the same issue of Science addressing the implications of the research.
ANALYSIS The regulatory gap is concrete: existing federal policy targets gain-of-function research on natural pathogens, while the Stanford work produced organisms with no natural precedent — a category the current framework does not directly address.
The demonstrated ability to generate functional viral genomes from large language models trained on DNA marks a qualitative expansion of AI's role in biology, moving from protein design to the creation of complete, viable organisms.