Overview
- A Stanford and Arc Institute team reported in Science that generative AI produced complete bacteriophage genomes that were synthesized and produced functioning phages in lab assays on August 6.
- The experiment targeted bacteriophages that infect bacteria only and the researchers removed human, animal, plant and fungal virus sequences from training data while confining wet‑lab work to E. coli.
- The team used genome‑scale models developed at the Arc Institute (reported as Evo 1 and Evo 2), generated hundreds of candidate genomes, synthesized almost 300 designs, and validated 16 that infected and killed E. coli in culture.
- Biosecurity experts and one of the model developers warned the technique can produce functionally equivalent sequences that may evade existing DNA‑synthesis screening and immune detection, creating so‑called biological 'deepfakes' and novel dual‑use risks.
- Researchers note the work could speed tailored phage therapies for antibiotic‑resistant bacteria but said clinical safety testing, technical scaling to larger genomes, stronger DNA‑order screening, and new governance are needed before medical use.