Overview
- Researchers at Stanford and the Arc Institute published a Science paper on Friday that reports AI models named Evo1 and Evo2 produced complete genomes which yielded 16 synthetic bacteriophages able to kill E. coli strains resistant to natural phages.
- The team trained the genomic language models on about two million bacteriophage sequences, generated thousands of candidate genomes, selected 302 for DNA synthesis and lab testing, and found 16 viable phages.
- The synthetic phages have very small genomes of roughly 5,400 base pairs, infect only bacteria and have no demonstrated ability to infect human cells.
- The authors explicitly warned the work raises 'important urgent' biosafety questions and urged consultation with security professionals, and experts cautioned the same tools could in principle be misused even though creating dangerous human pathogens remains illegal and hypothetical today.
- If developed carefully, the method could speed tailored phage therapy for antibiotic resistance and change how scientists design antimicrobial agents, but it also heightens calls for clearer governance, independent review, and rules for safe use of genome‑design AI.