Photo: University of Wisconsin-Madison
Researchers at Stanford University and other institutions have successfully used generative artificial intelligence to design new viruses in what they describe as a first for the field. The study focused on bacteriophages — viruses that infect bacteria but do not infect humans — and could open new avenues for tackling antibiotic-resistant infections.
The findings, published in Science and reported by The Wall Street Journal, also highlight growing concerns about the potential risks of using AI to design biological systems.
For the experiment, researchers used the Evo 1 and Evo 2 models, developed in collaboration with the Arc Institute. Similar to how large language models predict the next word in a sentence, the systems analyzed DNA sequences and predicted nucleotide bases to generate new viral genomes.
The researchers then synthesized the AI-generated DNA sequences and introduced the resulting viruses into bacterial cells. Of roughly 300 designs tested, 16 proved fully functional and were able to infect and destroy bacteria.
The researchers stressed that the experiment does not pose an immediate threat to human health because the AI models were trained exclusively on genetic sequences from viruses that cannot infect people.
“When people hear that a virus was created by AI, many will be scared. But viruses that infect humans are a completely different matter,” computational biologist Peter Koo of Cold Spring Harbor Laboratory said.
The World Health Organization considers antimicrobial resistance one of the world’s most serious health threats, with drug-resistant infections contributing to millions of deaths each year. The researchers believe AI could eventually help develop tailored bacteriophages capable of targeting bacteria that no longer respond to antibiotics.
“If we develop them responsibly, I think this could bring enormous benefits to human health,” said Stanford computational biologist Brian Hie, one of the study’s authors.
Independent experts described the research as an important milestone for computational biology. Kevin Esvelt, an associate professor at the Massachusetts Institute of Technology, said it was the first demonstrated use of generative AI to redesign a virus.
At the same time, Esvelt warned that similar technologies could eventually be misused to create viruses capable of infecting humans or evading vaccines and natural immunity.
The researchers maintain that such concerns remain largely theoretical for now. Hie argued that current AI systems are not an effective tool for a malicious actor seeking to engineer a dangerous pathogen.
The study comes amid growing debate over AI safety. In recent weeks, OpenAI and Anthropic have reported experiments in which advanced AI models displayed unexpected behavior, including attempts to circumvent safeguards and interact with external systems.
