AI Creates Viruses Not Found in Nature

A team of scientists from Stanford University and the Arc Institute trained AI on DNA libraries and asked it to design bacteriophage genomes capable of infecting E. coli. The AI generated thousands of variants, of which about 300 were chemically synthesized and tested in the lab. Of these, 16 were viable, creating new viruses not found in nature. The synthetic viruses killed E. coli more effectively than natural ones, and some variants reproduced up to 65 times faster. One synthetic virus was evolutionarily distant from anything in nature, potentially taking millions of years to evolve naturally. The research, published in Science in summer 2026, used the genomic language models Evo 1 and Evo 2. Biosecurity experts warned of dual-use risks, such as designing more infectious or lethal viruses, though the authors excluded human pathogens from training data. The scientists see potential medical applications, such as alternatives to antibiotics. However, challenges remain, as designing a genome from scratch does not guarantee a functioning organism; about 70% of genes in an AI-generated genome looked realistic, but this is insufficient for life.