AI designs viruses never seen in nature

Stanford and Arc Institute researchers achieved a significant milestone by using AI language models (Evo 1 and Evo 2) to design 16 completely new viruses never found in nature. The models were trained on millions of existing genomes and tasked with creating variants of Phi X174, a bacteriophage that infects E. coli. Of 285 synthesized phages tested, 16 were viable, with some replicating faster than the original virus and several distinct enough to qualify as new species. The AI-designed viruses successfully eliminated E. coli strains that had developed resistance to the natural version, demonstrating practical therapeutic potential for treating antibiotic-resistant infections.

The research raises significant biosecurity concerns despite safety measures taken during development. The researchers deliberately avoided training the models on viruses that infect humans, animals, or plants to prevent generation of dangerous pathogens. However, the technology is fundamentally dual-use: the same AI capabilities that design beneficial bacterial viruses could theoretically be repurposed to create harmful pathogens if trained on different datasets. Evo 2 has been released as open source, and the field is advancing rapidly, creating pressure for the development of biosafety testing frameworks and regulatory guardrails.

Anthropically addressed related concerns by rewriting the safety classifier for their Fable 5 model, which handles biology-related queries. The original system was overly restrictive, blocking nearly all biology questions even from legitimate researchers. The updated classifier reduces false positives by approximately 85%, allowing the model to answer educational and healthcare questions while still restricting dual-use research involving virology, toxicology, and molecular design. This represents a deliberate balance between enabling beneficial access and preventing misuse, though the company acknowledges it may frustrate legitimate researchers seeking capabilities for important work.