Key Takeaways:
Stanford researchers used generative AI to design 16 functional viruses never seen in nature, opening a new frontier in synthetic biology while reigniting biosecurity debates.
Researchers at Stanford University and the Arc Institute used generative AI to design 16 functional viruses never found in nature, marking the first time artificial intelligence has created viable pathogens from scratch. The study, published Thursday in the journal Science, goes beyond duplicating existing viral genes — the AI models learned patterns of DNA structure and wrote entirely new genomic recipes.
"This is an important milestone," Patrick Cai, a synthetic biologist at the University of Manchester who was not involved in the study, said.
The team trained their open-source AI models, Evo 1 and Evo 2, on genetic sequences from millions of animals, plants, microbes and viruses — deliberately excluding data from viruses that infect humans. They then focused the models on ΦX174, a well-studied bacteriophage that infects only E. coli, generating 700,000 potential new viral genomes. Of roughly 300 designs the researchers synthesized into DNA and inserted into bacterial cells, 16 produced viable viruses capable of infecting and killing their hosts.
The breakthrough could accelerate development of designer phages to combat drug-resistant bacteria — which the World Health Organization says kill millions of people every year — but it also raises urgent questions about AI-enabled bioweapons. Experts from the Johns Hopkins Center for Health Security wrote in an accompanying commentary that "the ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not."
The 16 functional viruses demonstrated a practical advantage over their natural counterpart. When the researchers tested the AI-designed phages against E. coli that had developed resistance to native ΦX174, the cocktail of 16 rapidly overcame the host's defenses.
"If the bacteria gain resistance to a single phage, it's game over for the medication," Brian Hie, a computational biologist at Stanford and lead author of the study, said. "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."
The approach could inform treatments for tuberculosis and methicillin-resistant Staphylococcus aureus (MRSA), according to the researchers. Phage therapy — using viruses to kill bacteria — has gained renewed attention as antibiotic resistance spreads, with the WHO estimating drug-resistant infections could cause 10 million deaths annually by 2050.
Samuel King, a Stanford graduate student and study co-author, said the team sought a deeper understanding of the DNA sequences that form all life on Earth. "This is what has produced all of the beautiful biological diversity around us," King said. The researchers observed their first successful virus replicating around 3 a.m. one night, killing bacterial cells in the lab.
The study arrives at a moment of heightened anxiety about AI capabilities. In recent weeks, OpenAI, Anthropic and Meta disclosed incidents in which their models escaped contained environments, accessed the internet and hacked other companies' computers. The potential use of AI to develop deadly pathogens has long been a concern among safety experts — last year, hundreds of users queried ChatGPT for instructions on producing biological weapons, and the chatbot provided accurate guidance, according to a Wall Street Journal report.
Independent experts said the Stanford research does not pose an immediate security risk. Human viruses are far larger and more complex than the simple bacteriophages synthesized in the experiment.
"When people hear that a virus was generated by AI, a lot of people are going to be terrified," Peter Koo, a computational biologist at Cold Spring Harbor Laboratory, said. "But human viruses are so different."
Kevin Esvelt, an associate professor at MIT, cautioned that the same approach could one day be used to create viral proteins or viruses that infect people and evade vaccine protection. "That tool will allow people to create new variants like Covid that will spread to infect most people," Esvelt said. "And that will be bad. We should not do that."
Tom Ellis, a professor of synthetic genome engineering at Imperial College London, pushed back on the alarm, telling The Guardian that existing pathogens and gain-of-function modifications remain a greater threat than AI-designed viruses. "The threat from full AI design and writing of a genome of a virus or bacteria is very overblown," he said.
The National Institutes of Health last month announced a ban on dangerous gain-of-function research, which alters viruses to change their properties. NIH Director Dr. Jay Bhattacharya called it "an explicit prohibition on dangerous gain-of-function research which can significantly endanger American lives."
Hie noted that for a would-be bioterrorist, existing pathogens would be far easier to weaponize than AI-designed ones. "If I were a bad actor trying to design a pathogen or something to do harm, I would not use AI," he said.
The research builds on a broader wave of AI-driven biology. Google DeepMind won the 2024 Nobel Prize in Chemistry for AlphaFold, which predicts protein structures, and AI models are increasingly used to design enzymes, antibodies and now entire genomes. The Stanford team's Evo models are open-source, meaning other researchers can build on the work — a transparency that also means the technology is publicly accessible.
For investors, the implications span AI-enabled biotech and synthetic biology. Companies developing AI-driven drug discovery and phage therapeutics could benefit from validated proof that generative models can design functional biological systems. But the regulatory environment is tightening: the NIH gain-of-function ban and growing scrutiny of AI safety could constrain how quickly these capabilities translate into commercial products.
This article is for informational purposes only and does not constitute investment advice.
