Overview
- Researchers at Stanford and the Arc Institute used generative models called Evo 1 and Evo 2 to produce roughly 700,000 candidate DNA sequences and then synthesized about 285 for laboratory testing.
- Laboratory work converted 16 AI‑designed genomes into functioning bacteriophages that replicated in a non‑pathogenic strain of E. coli, and the study was published in Science.
- The team deliberately restricted the work to phages, excluded animal and plant viruses from training, and said it disabled parts of the model to prevent designs for human‑infecting viruses.
- Evo 2 is available as open source, which has led biosecurity experts to call for stronger rules and clearer oversight because public access can complicate control of potentially risky tools.
- Authors say the method could help develop therapies against antibiotic‑resistant bacteria, but experts note technical barriers remain for making human‑infecting agents and warn the finding exposes regulatory gaps for computational genome design.