Stanford scientists used the AI model Evo 2 to design 16 viruses that destroy antibiotic-resistant bacteria, paving the way for a new generation of drugs, but also sparking debate over biosecurity.
Scientists at Stanford University have achieved a historic breakthrough by using generative artificial intelligence to create the first viruses specifically designed to combat drug-resistant superbugs. The team, led by Brian Hie, a chemical engineer and co-author of the study, used an AI model called Evo 2, trained on millions of genetic sequences from around the world, to create a set of viruses known as bacteriophages-microorganisms that hunt down and kill harmful bacteria.
In laboratory tests, a cocktail of 16 "exceptionally good" viruses designed by Evo 2 was able to rapidly destroy E. coli bacteria that were already immune to natural phages. E. coli is a group of bacteria that causes intestinal infections and is increasingly resistant to available antibiotics, making growing antibiotic resistance a global threat.
The key advantage of this approach lies in combining multiple genetically distinct viruses. "If bacteria develop resistance to one phage, the drug becomes useless," Brian Hie explained to Euronews. "But if you have multiple genetically distinct phages in a cocktail, it will be significantly harder for bacteria to develop resistance to the entire mixture."
The researchers note that this technology could potentially be used in the future to target other dangerous bacteria, such as those causing tuberculosis or the common hospital infection MRSA.
While the discovery offers immense medical potential, the fact that the researchers made the Evo 2 model openly and freely available for download has sparked serious safety debates. The study authors acknowledge that the tool's openness has raised concerns that "malicious actors" could potentially use modified versions to design harmful biological agents.
Simon Clarke, an associate professor of cellular microbiology at the University of Reading in the UK, who was not involved in the research, commented to Euronews: "As the authors point out, this raises serious regulatory and safety questions, to say the least."
"While work of this nature is usually strictly regulated, it is encouraging that these scientists have shown additional restraint by providing important safeguards, but there is no guarantee that every other scientist attempting something similar will be equally cautious," Clarke added.
The team behind the Evo 2 model has taken concrete steps to mitigate risks. As early as February 2025, they announced that they had excluded pathogens that infect humans and other complex organisms from their datasets, citing ethical and safety risks and their desire to "prevent the use of Evo for developing biological weapons."
The study authors argue that the advantage of AI-designed biology over natural evolution is precisely the ability to directly incorporate safety checks into the process itself. Clarke, however, argued that naturally occurring pathogens currently pose a greater risk than those designed by AI because they are easier to obtain and produce than creating new ones from scratch.
The Evo 2 model was trained on millions of natural genomes, allowing it to learn the complex "grammar" and rules that make a DNA sequence functional. As with other biological AI models, this dataset includes genetic sequences and characteristics of pathogens.
Currently, there is no universal framework regulating these datasets. While some developers voluntarily exclude high-risk data, researchers argue that clear and consistent rules should apply to everyone. Earlier this year, more than 100 researchers worldwide wrote an open letter stating: "The stakes of biological data governance are high because AI models could help create serious biological threats."
The scientists concluded that finding the right balance between openness and necessary safety restrictions for high-risk data will be crucial as AI systems become more powerful and widely available.
