AI learns to create new viruses: a breakthrough in medicine or a biosecurity threat?

A breakthrough has occurred in the world of biotechnology and artificial intelligence that could radically change the approach to treating bacterial infections. According to a study published in the prestigious journal Science, scientists have demonstrated the ability of AI to construct the genomes of new viruses. This technology, based on specialized language models, opens the way to creating "smart" bacteriophages — viruses that selectively destroy pathogenic bacteria without affecting healthy cells.

How genetic AI works: from nucleotides to new viruses

The development is based on the Evo 1 and Evo 2 models. Unlike familiar chatbots trained on text and web content, these systems "study" the language of life — trillions of nucleotides, which are the building blocks of DNA. Researchers tasked the AI with mastering genetic "grammar" so it could independently assemble functional genomes. To train the model, they used a database of 15,000 viruses from the Phi X-174 family — microscopic viruses that infect only E. coli. The result was the creation of viruses capable of infecting bacteria that were previously not susceptible to this type of infection.

Medical potential: a future without antibiotics?

Scientists emphasize that the main goal of the development is to accelerate the creation of therapies for dangerous bacterial infections. In the context of growing bacterial resistance to antibiotics (superbugs), the ability to quickly design killer viruses becomes critically important. The technology allows for the creation of tools for targeted attacks on pathogens in a controlled environment, which could be a lifesaver for patients with incurable infections. This opens the era of personalized phage therapy, where treatment is tailored to a specific bacterial strain.

Controversial data

Despite the researchers' optimism, serious controversies have arisen around the technology. On the one hand, the authors of the study claim that they consciously limited the AI parameters, excluding any structures capable of infecting humans, animals, plants, or fungi. On the other hand, biosecurity experts point to a fundamental problem: if basic chatbots are already capable of providing instructions for creating biological weapons, then specialized models at the level of Evo could be used by malicious actors to create highly contagious or lethal viruses. The risk lies in the fact that the "genetic grammar" learned by the AI could be retrained to create threats that go beyond medical purposes.

Regulatory vacuum and challenges of 2026

The situation is exacerbated by the fact that the pace of AI technology development is already outpacing current biosecurity norms. To date, there are no strict international barriers that could control access to such gene design tools. Experts are calling for the urgent creation of control mechanisms, as the ability to design genetic code becomes more accessible. In 2026, the question of who and how will regulate the creation of synthetic viruses is becoming one of the key issues for global security.