Scientists at Stanford University have used artificial intelligence to design entire viral genomes for the first time. The AI-designed viruses successfully replicated in laboratory conditions, according to a study published in Science.
Researchers employed generative models Evo1 and Evo2, trained on the genetic codes of viruses, bacteria, plants, and humans. The technology functions similarly to large language models like ChatGPT but predicts DNA sequences instead of words.
The machines were refined to generate bacteriophages—viruses that infect specific types of bacteria.
From thousands of generated variants, scientists selected 302 of the most promising and synthesized them in the lab. Sixteen of these proved effective against E. coli, with some viruses overcoming the bacteria’s natural defense mechanisms.
The created sequences exhibit patterns distinct from anything found in nature.
Graduate student Samuel King, who participated in the experiment, told BBC that the team noticed the phages’ activity early in the morning: clear spots appeared on Petri dishes with bacteria, indicating viral replication. When the results were shown to the entire group, “the room erupted in applause,” noted Stanford associate professor Brian He.
Scientific Community Reaction
The development of new bacteriophages could pave the way for treating antibiotic-resistant infections. However, the publication sparked a debate about safety.
In an accompanying commentary in Science, Thomas Inglesby and Moritz Hanke from the Johns Hopkins Center for Health Security stated that the results raise “urgent biosecurity questions.”
“You could say, ‘Hey, genomic language model, make me a flu genome modified to be more contagious or more lethal,'” warned Hanke.
Experts stated that creating new viruses capable of causing human diseases “should not be pursued.”
The researchers themselves took precautions: viruses affecting complex organisms were excluded from the training base, and work was conducted with bacteriophages in a secure laboratory. He noted that existing safety measures largely ensure the technology will be used for good.
However, not everyone shares the concern. Tom Ellis from Imperial College London emphasized that the AI-created viruses are “literally the smallest and simplest genomes.” According to him, creating a pathogen dangerous to humans would require significantly more complex work.
Marc Güell from the Synthetic Biology Laboratory at Pompeu Fabra University in Spain called the research a “very significant turning point,” as “for the first time in history, people are beginning to design biology on a computer.”
He noted that viruses are not living organisms, and creating living cells with AI would require another significant leap: a phage genome contains about 5,400 base pairs, while the minimal genome of a living cell is about 500,000, and the human genome is three billion.
In July, AI startup Anthropic launched the Claude Science environment for scientists and announced plans to develop treatments for “advanced” diseases.
