AI Generates Functional Viruses, Sparking Dual-Use Safety Debate

The study, published in the journal Science, utilized an AI system dubbed Evo, which was trained on nine trillion genetic sequences from viruses, animals, and plants. Unlike models designed for language, Evo identifies patterns within biological code to synthesize entirely new genetic instructions. By targeting Phi X-174—a virus infecting E. coli—the team tested the AI’s ability to move beyond merely copying known genetic material to drafting original, viable structures.

After Evo generated approximately 700,000 potential designs, researchers synthesized 285 candidates to test in a laboratory setting. Sixteen of these designs proved functional, successfully infecting and killing bacteria in petri dishes. While these specific viruses pose no risk to human health, the experiment validates a methodology that could theoretically be applied to more complex or harmful organisms. Patrick Cai, a biologist at the University of Manchester, noted that the achievement marks a significant milestone in synthetic biology, though it concurrently intensifies the urgency for safety protocols as generative tools become increasingly adept at manipulating the building blocks of life.