In a breakthrough that could reshape the fight against antibiotic-resistant bacteria, scientists have engineered the first synthetic viruses designed entirely by artificial intelligence. However, while the development offers renewed hope for treating intractable superbug infections, it has simultaneously ignited intense debate over the biosecurity risks of generative biology.
Researchers at Stanford University, led by chemical engineer Dr. Brian Hie, utilized advanced genome language models to construct functional genetic code from scratch. The models, named Evo1 and Evo2, were trained on a database of two million bacteriophages—specialized viruses that attack bacteria without harming human cells. To mitigate biosecurity hazards during training, genetic data from viruses targeting humans, animals, and plants was deliberately omitted.
From thousands of AI-generated candidate sequences, the team synthesized nearly 300 prospective genomes in the laboratory. When introduced into bacterial hosts, 16 viable bacteriophages emerged. In laboratory trials detailed in the journal Science, a tailored cocktail of these artificial viruses successfully destroyed drug-resistant strains of E. coli that had resisted natural phages.
The capability to rapidly tailor synthetic phages could revolutionize phage therapy, providing a nimble weapon against mutating superbugs. Yet experts warn that the milestone exposes glaring gaps in global oversight. Writing in an accompanying commentary, biosecurity specialists Prof. Tom Inglesby and Dr. Moritz Hanke from Johns Hopkins University cautioned that while the capability to generate viral genomes now exists, the regulatory framework to safely steer the technology remains unformed.
Biosecurity analysts stress that safeguards must evolve alongside the technology. While some researchers point out that modifying existing pathogens remains a more immediate danger than designing new ones from scratch, experts such as Dr. Filippa Lentzos of King's College London advocate for a layered defense system. Such measures include rigorous DNA synthesis screening, access restrictions on generative models, and updated laboratory biosecurity standards to ensure that computational tools do not accidentally unlock catastrophic biological risks.