For the first time, scientists have used artificial intelligence to create new kinds of viruses, a development that could speed up medical research while raising concerns about the potential for AI-designed pathogens. Researchers at Stanford University and the Arc Institute in Palo Alto, Calif., trained an AI model on libraries of DNA and asked it to write recipes for viral genomes. Sixteen of the resulting viruses were viable.
Synthesizing viruses from scratch is not new. Researchers have long manufactured viral genomes to study antiviral drugs and vaccines and to learn how viruses work. The new study, published Thursday in the journal Science, goes further: it uses AI to invent genomes that have never existed in nature.
How the AI designed the viruses
The researchers taught the AI to recognize patterns in natural DNA structure. They then asked the model to generate recipes for new viral genomes. The team synthesized the resulting DNA in the lab and inserted it into bacteria. The modified bacteria produced viruses, and 16 of those viruses could infect other bacteria, showing they were fully functional.
The AI-generated viruses all resemble Phi X-174, a naturally occurring virus that infects only bacteria, so none of them pose a threat to humans. Patrick Cai, a synthetic biologist at the University of Manchester who was not involved in the study, said: "This is an important milestone."
What the results mean for biosafety
The ability to produce viable viruses from AI-designed genomes opens new possibilities for studying viral evolution and testing countermeasures. It also highlights a risk: the same approach could eventually be used to create dangerous pathogens. Because the technology is still new, oversight and responsible use will be central questions as it develops.
Why this matters for science and research professionals
For researchers in virology, genomics, and synthetic biology, this study shows that AI can do more than analyze data - it can generate functional biological designs. Scientists who want to build these skills can find structured training through AI for Science & Research and AI Learning Path for Research Scientists. Understanding how AI models learn biological patterns will be increasingly important for anyone working at the intersection of computation and biology.
Your membership also unlocks:
