AI Created 16 Functioning Viruses from a Pool of 700,000 Candidates

Arc Institute + Stanford team has managed to synthesize 16 functional bacteriophages using their genomic language models Evo 1 + Evo 2. The models were trained on trillions of nucleotides and 15,000 viruses. Out of 700,000 candidate genomes, 285 were tested and 16 of them were found to be functional (some of them even replicate faster than the original). They are designed to attack E. coli only — there’s no way they can be used to hurt humans intentionally.

In this article in Science, the authors pose an interesting question that regulators should consider over the next decade: if a model can describe pathogens, it provides information, but if it can generate genomes, it wields creativity. This is where we need to draw the line between biosecurity.