Using AI, scientists have created viruses to combat superbugs
Scientists have, for the first time, created viable bacteriophages whose genomes were entirely designed by generative artificial intelligence. The new viruses were able to destroy strains of E. coli that are resistant to natural bacteriophages. At the same time, researchers warn that while the technology opens up new possibilities for medicine, it also raises serious questions about biosafety.
The study was conducted by a team led by Stanford University chemical engineer Brian Gui. The scientists used two AI models—Evo1 and Evo2. These models analyze genetic sequences in much the same way that language models process text. Genetic data from approximately two million bacteriophages were used to train the AI.
Bacteriophages are viruses that infect and destroy bacteria. They do not affect humans. Such viruses are already used in phage therapy, particularly to combat bacterial infections when antibiotics are no longer effective. The researchers deliberately excluded from the training data the genomes of viruses capable of infecting humans, animals, or plants. In this way, they sought to reduce the risk that the system would create a dangerous pathogen.
The AI proposed thousands of variants
After training the model, it generated thousands of possible bacteriophage genomes. The scientists selected nearly 300 variants and attempted to recreate them in the laboratory. To do this, they inserted the genetic material into bacteria. The bacteria read the resulting code and began producing new bacteriophages.
The experiment was not successful with all variants. Only 16 of the created viruses proved to be viable. However, this proved sufficient to achieve an important result. A mixture of the new bacteriophages was able to quickly overcome the resistance of two different strains of E. coli.
How This Could Help Medicine
According to the study’s authors, this technology could revolutionize the approach to phage therapy. If scientists can rapidly create bacteriophages tailored to specific bacteria, this could potentially enable a more rapid response to infections that no longer respond to available treatments.
In other words, instead of a lengthy search for a natural virus against a specific bacterium, in the future it may be possible to design the necessary bacteriophage using AI. At the same time, this is currently just a promising scientific approach, not a ready-made treatment. The technology’s effectiveness and safety still require further research.
Scientists Warn of Risks
Along with its potential benefits, the research raises complex questions about biosafety. Scientists emphasize that the ability to create complete viral genomes using AI already exists, but a regulatory framework for such technologies has not yet been established. Professor Tom Inglesby and Dr. Moritz Hanke of the Johns Hopkins University Center for Health Security have drawn attention to this very issue.
“The ability to create viral genomes using generative AI already exists, but there is still no regulatory framework in place to safely guide this technology,” they noted. The researchers emphasized that this approach should not be applied to viruses capable of infecting humans, animals, or plants. Pathogens created in this way could theoretically possess properties that would make it difficult to combat them using existing vaccines, medications, or other treatments.
Is It Possible to Create a Dangerous Virus?
Tom Ellis, a professor of synthetic genomics at Imperial College London, called the study’s results impressive but urged people to consider the scale of the experiment. According to him, bacteriophages have very small and simple genomes. Therefore, creating such viruses using AI is much simpler than designing more complex organisms. “It’s literally the smallest and simplest genome you can create,” Ellis explained.
He also noted that the threat posed by pathogens created entirely by AI should not be exaggerated. According to the expert, it is still significantly easier today to modify existing dangerous organisms than to create a complex pathogen from scratch. At the same time, risks can be mitigated by controlling access to genetic data and verifying orders for the synthesis of suspicious DNA sequences.
It’s Not Just AI That Needs to Be Monitored
Philippa Lentzos, an expert in science and international security at King’s College London, believes that special attention should be paid to the moment when a designed genetic sequence is physically created in a laboratory. According to her, protection must extend beyond the AI models themselves. It is necessary to control access to such systems, evaluate research projects, verify genetic sequences before their synthesis, and ensure compliance with laboratory safety standards.
Thus, the experiment to create bacteriophages revealed two sides of this new technology at once. Artificial intelligence is already capable of helping to create viable viruses that have the potential to combat drug-resistant bacteria. At the same time, however, the scientific community must determine how to prevent these capabilities from being used to create dangerous pathogens.
