Researchers at Stanford University have achieved a groundbreaking milestone by utilizing artificial intelligence to design complete viral genomes. The AI-generated viruses were successfully cultivated in laboratory settings, as detailed in a study published in the journal Science here.
The scientists employed generative models named Evo1 and Evo2, which were trained on the genetic codes of viruses, bacteria, plants, and humans. This technology operates similarly to large language models like ChatGPT, but instead of predicting words, it forecasts DNA sequences.
These models were specifically enhanced to create bacteriophages—viruses that target particular bacterial species.
From thousands of generated variants, the team selected 302 promising candidates and synthesized them in the lab. Among these, 16 demonstrated effectiveness against Escherichia coli (E. coli), with some successfully overcoming the bacteria's natural defenses.
The sequences created exhibit patterns not found in natural organisms.
Graduate student Samuel King, who participated in the experiment, shared with BBC that the team observed the phages’ activity early one morning when they noticed clear spots on Petri dishes containing bacteria—signs of viral replication. When the results were presented to the entire group, “the room erupted in applause,” noted Stanford associate professor Brian Hays.
Reactions from the Scientific Community
The development of new bacteriophages has the potential to pave the way for treating antibiotic-resistant infections. However, the publication has sparked discussions regarding safety concerns.
In a commentary accompanying the article in Science, Thomas Inglesby and Moritz Hanke from the Johns Hopkins Center for Health Security stated that the findings raise "urgent questions about biosecurity."
“You could say, ‘Hey, genomic language model, make me a flu genome modified to be more contagious or more lethal,’” warned Hanke.
Experts have asserted that creating new viruses capable of causing human diseases should not be pursued.
The researchers themselves took precautions by excluding viruses that infect complex organisms from their training dataset and conducted their work with bacteriophages in a secure laboratory. Hays emphasized that existing safety measures largely ensure that the technology will be used for beneficial purposes.
Nevertheless, not everyone shares these concerns. Tom Ellis from Imperial College London stated that the AI-generated viruses are “literally the smallest and simplest genomes.” He pointed out that creating a pathogen dangerous to humans would require a significantly more complex process.
Mark Guell from the Synthetic Biology Lab at Pompeu Fabra University in Spain described the study as “a very significant turning point,” noting that “for the first time in history, humans are beginning to design biology on computers.”
Hays added that viruses are not considered living organisms, and creating living cells using AI would require another substantial leap: a phage genome contains about 5,400 base pairs, while the minimum genome of a living cell is around 500,000, and a human genome comprises three billion.
It is worth mentioning that in July, AI startup Anthropic launched a platform for scientists called Claude Science, announcing plans to develop treatments for “advanced” diseases.