Artificial Intelligence Creates Novel Viable Viruses Not Found in Nature

Researchers at the Arc Institute in Palo Alto, USA, have for the first time used artificial intelligence to create fully viable bacteriophages not found in nature—viruses that destroy bacteria as their targets. The scientific experiment, published in Science, could mark a major breakthrough in nanotechnology and biomedicine, while also raising serious safety concerns among experts. This was reported by Ixbt.com .

According to ixbt.com, scientists trained a specialized genome language model called Evo on millions of DNA sequences from all domains of life to generate new genomes. Based on the principles of large language models designed for text, the system identified the evolutionary constraints governing natural genomes and enabled researchers to design entirely new variants. During the study, the Phi X-174 bacteriophage, harmless to humans and capable of infecting Escherichia coli, was used as the main template.

Hundreds of Thousands of Genomes Created by Artificial Intelligence

According to The New York Times, the Evo model generated 700,000 potential genomic sequences during testing. Researchers then selected the 300 most promising candidates, synthesized them in the laboratory, and inserted them into bacteria. Mature virus particles formed from these candidates, and 16 phages were proven to be fully viable and active.

The synthesized phages differed significantly from their natural counterparts in structure. They contained entirely new mutations, unusual genes, and regulatory elements, and also varied in length. For example, one phage used a DNA-packaging protein in its capsid—the protein shell—that belonged to another evolutionarily distant virus.

Fighting Antibiotic-Resistant Bacteria

Laboratory tests showed that the AI-created phages had varying levels of infectious activity and could even infect E. coli strains that had developed resistance to the natural Phi X-174 phage. A mixture of several generated phages fought resistant bacteria very quickly, while a mixture of natural phages was ineffective. This suggests that the methodology could become an important step toward developing effective phage therapy against antibiotic-resistant bacteria.

Biosafety and Regulatory Challenges

However, the technology’s rapid development is causing legitimate concern among experts. In a co-authored article in Science, Tom Inglesby, a professor at the Johns Hopkins Center for Health Security, and Dr. Moritz Hanke noted that the ability to synthesize virus genomes using generative artificial intelligence had emerged ahead of systems for controlling it. In their view, mechanisms for safely directing the technology do not yet exist.

Although the US National Institutes of Health (NIH) has introduced new rules banning experiments that increase the risk posed by biological agents, the restrictions do not extend to computational research. The document permits the modeling of new biological forms as long as they are not based on existing dangerous pathogens, such as the smallpox virus. As a result, the development of genome generators using artificial intelligence currently remains outside the scope of direct government regulation.