Researchers at Stanford University used artificial intelligence to create 16 functional viruses not found in nature, a move that could point toward new treatments for antibiotic-resistant infections. The viruses, called bacteriophages, target bacteria exclusively, and a mixture of the lab-made phages was able to kill E. coli strains that had developed resistance to naturally occurring phages.
Chemical engineer Brian Hie and Samuel King led the work, using EVO 2, a generative AI model Hie created that produces new DNA sequences. Using the phage ΦX174, which infects E. coli, as a template, the AI generated thousands of new genomes. The team chemically synthesized and tested nearly 300 of them, and 16 proved viable and effective - all distinct from any natural phages.
Antibiotic-resistant E. coli is a real threat in Israel. A 2021 study by the Health Ministry and Tel Aviv University found the bacteria are placing an "extremely high" burden on the public health system and called for prevention strategies and long-term control of antibiotic resistance.
The Canadian researchers think a mixture of these designed phages could lead to better antibiotics than single-phage treatments. "If the bacteria gain resistance to a single phage, it's game over for the medication," Hie said to the Stanford Report. "But if you have multiple genetically distinct phages in a mixture, it would be harder for the bacteria to develop resistance to the entire cocktail."
Safety concerns and open access
The AI model is open source, so anyone can download it and design new genomes. The Stanford researchers said they considered biosafety and wrote in their paper that all experiments were performed at the biosafety level appropriate for bacteriophage research.
They tried to limit risk by limiting EVO 2's training data, which means the model cannot generate viruses that target animals, plants, or fungi. Hie argued the benefits of open access outweigh the risks, saying designed viruses were subject to more oversight in production than many naturally occurring viruses.
Scientists push for more oversight
Researchers at Johns Hopkins University's Center for Health Security, Principal Tresor Clause Inglesby and Moritz Hanke, disagree. In a written response, they said that while Stanford limited the AI's training data, other users could work around limitations and create viruses designed to infect humans, animals, or plants in ways we cannot counter.
Inglesby and Hanke are calling for national and global oversight, saying current mechanisms are not enough. Existing laws in the US and Israel govern research on known dangerous pathogens, but they don't cover new artificially designed viruses. International policies, such as WHO guidelines, also don't address AI-generated pathogens.
Stanford researchers suggested that providers of synthetic nucleic acids, used to create the viruses, should screen; Inglesby and Hanke say it should be mandated by law.
"The question is no longer whether generative viral genome design will exist. It is whether society can build oversight that allows its benefits to unfold while preventing it from enabling serious harm," they wrote.
Why this matters for Science and Research professionals
For researchers in the life sciences, this is the first clear demonstration that AI-designed biological tools can produce workable results with actual therapeutic potential - not just digital outcomes. The same open-source model that enables rapid discovery can be used for harm, creating new security review obligations for research institutions. The path from design to real-world application is on a scale of years, but the basic mechanism is now proven: AI can generate possibilities that then get chemically synthesized and tested. For those working on drug-resistant pathogen, this is a search that offers a library of new candidates. For those in research governance, this is a gap in the current framework. It's worth knowing which one you're in.
Your membership also unlocks:
