AI has, for the first time, designed complete viral genomes: a breakthrough that changes the game.
Stanford researchers have achieved a historic breakthrough: their generative artificial intelligence models have, for the first time, designed complete viral genomes that proved to be viable. The results, published in the prestigious journal Science, demonstrate that synthetic viruses successfully replicated under laboratory conditions.
How it works
At the core of the experiment are the Evo1 and Evo2 models, trained on vast arrays of genetic data—from viruses and bacteria to plants and humans. The operating principle resembles large language models like ChatGPT, but instead of predicting words, these algorithms generate DNA sequences. After refinement, the systems were tasked with creating bacteriophages—viruses that attack only specific types of bacteria.
From thousands of generated variants, specialists selected 302 of the most promising ones and synthesized them in the laboratory. The result is impressive: 16 of the created phages proved effective against Escherichia coli (E. coli), with some managing to overcome the bacteria's natural defense mechanisms. Of particular interest is the fact that the resulting sequences have no analogues in nature—these are fundamentally new biological structures.
A moment of triumph and safety concerns
Graduate student Samuel King recalls how the team discovered the phages' activity early in the morning: transparent spots appeared on the Petri dishes—a clear sign of viral replication. When the results were shown to the entire group, the laboratory erupted in applause. However, the scientists' euphoria is accompanied by serious concerns from biosecurity experts.
Thomas Inglesby and Moritz Hanke from the Johns Hopkins Center for Health Security emphasize that the technology raises "urgent biosecurity questions." Their main argument: theoretically, one could ask the model to create a modified influenza genome with increased transmissibility or lethality. In response, the researchers took preventive measures: viruses that infect complex organisms were excluded from the training dataset, and the work was conducted exclusively with bacteriophages in a secure laboratory.
However, not everyone shares the alarm. Tom Ellis from Imperial College London rightly notes that the AI-created viruses are "literally the smallest and simplest genomes," and creating a dangerous human pathogen would require disproportionately more complex work. Marc Güell from Pompeu Fabra University calls the study "a very significant turning point," emphasizing that humanity is beginning to design biology on a computer for the first time.
It is important to grasp the scale: a phage genome contains about 5,400 base pairs, the minimal genome of a living cell is roughly 500,000, and the human genome is three billion. This means the path from viruses to fully functional living organisms is still long, but the direction has been set.
My analysis: This breakthrough opens up enormous prospects for combating antibiotic-resistant infections, which is critically important amid the global crisis of resistance. However, it is also a reminder that AI is becoming a dual-use tool, and the regulation of such technologies must evolve in parallel with scientific progress, not after it.