Scientists at Stanford University and the Arc Institute created AI-generated viral genomes that produced 16 viable bacteria-infecting viruses, according to a study published in Science on August 6, 2026. A mixture of the viruses overcame antibacterial resistance in some E. coli strains—a promising experimental result that also prompted calls for stronger governance of biological AI.
Key takeaways
- Evo designed genomes for viruses targeting specific bacterial hosts.
- Laboratory tests yielded 16 viable viruses from about 300 designs.
- The resulting bacteriophages infect bacteria, not humans.
- Johns Hopkins specialists called for safeguards around biological AI models.
How Evo produced AI-generated viral genomes
The researchers used Evo, an AI program trained on genetic sequences from millions of sources across all domains of life. Those sequences served as its learning material, much as large collections of text underpin language models such as ChatGPT.
The team generated thousands of genome combinations within a framework designed for viruses that could use E. coli cells as hosts. Scientists then constructed and tested about 300 of those designs in the laboratory; 16 produced viable viruses.
According to the study, learning from the genetic library helped Evo recognize the evolutionary constraints that govern natural genomes. Yet the resulting designs differed from known natural viruses, and one contained a feature the researchers described as “evolutionarily distant.”
The finding suggested that AI had introduced changes that natural evolution might have taken millions of years to demonstrate were possible.
Bacteria-infecting viruses overcome resistance in tests
The viruses created in the experiment were bacteriophages: viruses that infect bacteria rather than humans. Their activity against resistant E. coli strains provided the study’s evidence of potential medical usefulness.
In testing, a mixture of the AI-designed viruses overcame antibacterial resistance in some E. coli strains. A comparable mixture of naturally sourced phages did not achieve that result.
The research points toward potential phage therapies against evolving pathogens, as scientists work to combat rising drug resistance in germs including E. coli. The reported result came from laboratory testing.
“The breakthrough achieved is significant,” said Jordi García Ojalvo, professor of systems biology at the Pompeu Fabra University of Barcelona, in a comment to the Science Media Centre.
Biosecurity specialists call for governance
Specialists at the Johns Hopkins Center for Health Security said AI-generated viral genomes raise urgent biosafety and biosecurity questions. In an accompanying Science article, they argued that governance has not kept pace with the ability to design viral genomes.
“The ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not,” they wrote.
The Johns Hopkins authors acknowledged that the research team addressed biosafety and biosecurity more deliberately than most developers of powerful biological AI models. They nevertheless singled out work on eukaryote-infecting pathogens as an area that should not be pursued, warning about designs that might infect humans, animals or plants beyond the reach of existing countermeasures.
García Ojalvo offered a more limited assessment of this experiment’s biosafety risk. He said it was lower than with some other AI tools because each designed genome requires individual testing and the proportion yielding viable viruses was low.
“It is difficult to imagine these models automatically generating viable genomes ‘out-of-the-box,’” he said.
Article produced with the assistance of artificial intelligence and reviewed by the editorial team.