AI-designed phages yield 16 viable genomes

Genome language models produced 16 viable bacteriophages after researchers generated thousands of designs and chemically synthesized nearly 300. The peer-reviewed Science paper reports that the Evo 1 and Evo 2 designs targeted E. coli C, differed from known natural phages, and included a mixture that overcame resistance where a comparable natural-phage mixture did not. For teams building four-day AI-to-wet-lab feedback loops, the operator lesson is governance rather than a recipe: whole-genome generation has crossed into experimentally viable output, so synthesis screening, contained validation, and auditable approvals must sit inside the workflow before throughput scales.