Artificial intelligence is rapidly changing how scientists study viruses by giving them tools to design both viruses and vaccines against them.
Researchers from Stanford University and the Arc Institute have used AI to design complete genomes for bacteriophages, which are viruses that infect bacteria rather than humans.
The team used Evo 1 and Evo 2, two genome language models trained on millions of natural genomes, to generate thousands of new genome designs using a known bacteriophage called ΦX174 as a starting point.
Researchers built and tested 285 of the designs in the laboratory, 16 of which produced functional bacteriophages capable of infecting and replicating inside E. coli, according to the study published in Science.
AI-designed viruses defeat resistant bacteria
The new phages were genetically different from known natural viruses and showed a range of structures and abilities.
Researchers combined several AI-designed phages into a cocktail that overcame resistance in three E. coli strains, succeeding where the natural ΦX174 virus could not.
The finding could support the development of phage therapies, which use bacteria-killing viruses to treat infections, including those that no longer respond to antibiotics.
However, the technology is still at an early stage. Only 16 of the 285 tested designs produced viable phages, and the researchers worked with one of the smallest and simplest viral genomes.
The models were prevented from generating viruses that infect humans, while the experiments used non-pathogenic bacteria under strict containment measures, the researchers said.
Still, experts writing in a related Science perspective warned that AI-designed viral genomes raise urgent safety and security questions, particularly if similar tools are eventually applied to viruses that infect humans, animals or plants.
AI harnesses the same design power for vaccines
While one research team used AI to design viruses, another used computational tools to develop a vaccine intended to protect against a broad group of coronaviruses.
Months before the virus study was published, AI had already helped scientists design a vaccine tested in humans, with researchers announcing the results of the 39-participant trial on June 5.
The vaccine was designed to target features shared by several related viruses, including SARS-CoV-2, SARS and bat coronaviruses that could potentially spread to humans.
Researchers reported no significant safety concerns and detected immune responses to several coronavirus targets. However, the findings remain preliminary, and further clinical trial phases involving larger and more diverse groups will be needed to assess the vaccine’s effectiveness.
AI’s growing biological power brings promise and risk
Both developments remain at an early stage and will require further research before they can be widely used, as the vaccine must pass larger clinical trials, while AI-designed phages will need extensive safety and effectiveness testing before they can be used to treat bacterial infections.
Together, however, the studies reveal the two sides of AI-powered biology. The technology could help scientists develop broader vaccines and new treatments for antibiotic-resistant infections, but its ability to design complete viral genomes also raises serious safety and security concerns.
As these systems become more advanced, their impact will depend not only on what scientists can create, but also on how carefully that power is tested, controlled and regulated.
