AI Creates First Viable Viruses That Never Existed
Scientists from Stanford and the Arc Institute have, for the first time, created viable artificial viruses using the AI model Evo. Trained on 9 trillion nucleotides, the model designed genomes capable of infecting bacteria. Of 285 synthesized sequences, 16 formed fully functional viruses, some of which replicated more actively than the natural prototype. The phages are safe for humans. The technology could help treat hereditary diseases, but experts fear regulatory frameworks are lagging behind.
Researchers at Stanford University and the Arc Institute have achieved a scientific first: using a specialized artificial intelligence model called Evo, they created viable artificial viruses. The model, trained on 9 trillion nucleotides, designed viral genomes capable of infecting and replicating within bacteria. In the experiment, based on the bacteriophage Phi X-174 and 15,000 similar viruses, Evo generated approximately 700,000 hypothetical genomes, from which scientists selected 285 for DNA synthesis. Of these, 16 formed fully viable viruses, with some strains replicating more vigorously than the natural prototype. The resulting phages are safe for humans, as pathogens were excluded from the training data. This technology could be applied to gene delivery for treating hereditary diseases. Biosecurity experts note that regulatory frameworks are struggling to keep pace with AI advancements, making it difficult to assess the potential dangers of novel genomes.