AI-Generated Viruses: A Breakthrough with Caution

Recent advancements in artificial intelligence have led to a significant breakthrough in the medical field, particularly in the development of therapies for persistent pathogens. Researchers have successfully utilized AI to create bacteriophages, which are viruses that specifically target bacteria, raising both excitement and serious safety concerns.

The study, published in the journal Science, highlights the potential of AI in creating effective treatments. Brian Hie, a chemical engineer at Stanford University and a co-author of the study, explained that his team used AI models named Evo 1 and Evo 2 to design functional genomes for bacteriophages. These genomic language models are comparable to the large language models used in chatbots like Claude, ChatGPT, and Gemini. The researchers trained these models on genetic data from two million bacteriophages while intentionally excluding the genetic codes for viruses that infect plants, humans, or other animals to mitigate safety risks.

The AI-generated thousands of potential genomes, from which nearly 300 were selected for laboratory synthesis. These genomes were introduced into bacteria, which then read the code and produced new bacteriophages. Although the process was not highly efficient, resulting in only 16 viable bacteriophages, the study indicated that a cocktail of these phages was effective in overcoming resistance in two different strains of E. coli. The researchers concluded that their approach paves the way for developing adaptive and resilient phage therapies against rapidly evolving pathogens.

Potential Risks Highlighted by Experts

Despite the promising developments, the researchers caution that this work raises significant safety questions. They urge other research teams to consult with experts during similar projects. In a companion article published in Science, Tom Inglesby and Moritz Hanke from the Center for Health Security at Johns Hopkins University echoed these warnings, stating, "While [the method] is promising for applications in the life sciences, it also raises urgent questions about biological safety and biosecurity. The ability to compose viral genomes using generative AI is now present; however, the frameworks to manage this safely are still lacking."

Inglesby and Hanke also noted that it remains unclear whether the same approach could be applied to other viruses. They cautioned against pursuing work on pathogens that could infect humans, animals, or plants. Tom Ellis, a professor of synthetic genomics at Imperial College London, also expressed concerns about the dangers. He warned that AI trained on the genetic codes of dangerous pathogens could potentially be used to design more harmful viruses. Nevertheless, he reassured the public, telling The Guardian that the risk of fully AI-generated virus or bacterium genomes is greatly exaggerated.