Scientists have used artificial intelligence to design 16 new viruses that no cell in nature has ever carried, a laboratory feat that could reshape how doctors fight drug-resistant germs and how governments police powerful biology tools. Researchers at Stanford University and the Arc Institute confirmed the new viruses in petri dishes, and their peer-reviewed report landed Thursday in the journal Science.
The work marks the first time a generative model has written full, working viral genomes end-to-end. Teams have long tapped software to design single proteins or small genetic parts. Now they can attempt entire genomes, so the new viruses push synthetic biology past a threshold that once seemed distant.
How did the models write the new viruses?
The researchers leaned on two genome language models, Evo 1 and Evo 2. These systems read DNA much as chatbots read text. Rather than guess the next word, they predict the next base pair. The team fed the tools roughly 2.7 million genomes drawn from across life, then prompted them to build fresh versions of ΦX174, a tiny bacteriophage that infects E. coli.
The models produced thousands of digital blueprints. Scientists then synthesized about 300, slipped the DNA into E. coli, and watched for signs of life. Only 16 came alive as working phages. That slim success rate matters. It proves the method works, yet it also shows the tools stumble far more often than they succeed.
Several of the new viruses did more than copy nature. A few outcompeted the natural ΦX174 phage in head-to-head growth tests. One design carried a DNA-packaging protein only distantly related to the known version. According to researchers, this is a breakthrough and an evolution that might have taken millions of years to achieve.
New viruses aim at drug resistance
The team also tested whether a mix of the new viruses could beat bacterial defenses. A cocktail of the designed phages quickly overcame ΦX174 resistance in three E. coli strains. A comparable blend of natural phages could not.
That result points toward phage therapy, a long-studied idea that turns viruses loose on bacteria antibiotics can no longer kill. The urgency is real. The World Health Organization reported in July that roughly one in six lab-confirmed bacterial infections worldwide resisted antibiotics in 2023. WHO also links bacterial resistance to more than 4.7 million deaths in 2021.
Still, the experiment stops well short of a cure. It stays a proof of concept, not a treatment for patients.
Jordi García Ojalvo, a systems biology professor at Pompeu Fabra University in Barcelona, praised the milestone while flagging its limits.
“The breakthrough achieved is significant,” García Ojalvo said.
Safety fears shadow the new viruses
The 16 new viruses infect bacteria, not people, and the team took steps to keep it that way. The models were never trained on sequences from pathogens that strike humans, animals, or plants. Even so, the power to compose viral genomes on a computer has rattled biosecurity experts. The researchers also released the fine-tuned models publicly, a choice that deepens the governance debate. Samuel King, the study’s lead designer and now a postdoctoral researcher, has said his group plans to keep building the phage program against antibiotic-resistant infections.
Researchers at the Johns Hopkins Center for Health Security raised their concern in a commentary that ran alongside the study.
“Although this is promising for life sciences applications, it also raises urgent biosafety and biosecurity questions,” they wrote.
They pressed the point further.
“The ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not,” they wrote.
The Johns Hopkins team drew one hard line. They urged scientists to steer clear of eukaryote-infecting pathogens, the kind that can sicken people, animals, and crops.
“Such genomes might encode new pathogens that can infect humans, animals, or plants in ways that cannot be contained by existing countermeasures,” they wrote.
Moritz Hanke, of the same center, sketched a darker scenario to The New York Times.
“You could say, ‘Hey, genomic language model, make me an influenza genome that is modified to be more transmissible or to be more lethal,’” Hanke said.
Other scientists urged calm. Tom Ellis, a synthetic genome engineering professor at Imperial College London, told The Guardian that building a human threat would not come easily. He called the phages “literally the smallest and easiest genome to make.”
García Ojalvo agreed the risk is modest at this time since each genome requires testing and the yield is low.
“It is difficult to imagine these models automatically generating viable genomes ‘out-of-the-box,’” García Ojalvo said.
For now, the 16 new viruses prove a point rather than pose a threat. Full-genome design has crossed from theory to the lab bench. The open questions are how far medicine should push these tools, and where regulators should draw the line.
What do you think? Could these new viruses become a weapon against drug-resistant infections, or do genome-writing tools carry dangers that outweigh the payoff? Please share your views below.