For the first time, artificial intelligence has designed functioning virus genomes from scratch, thrilling doctors and alarming biosecurity experts in equal measure.
Sixteen. That is the number that matters most in a new paper published in the journal Science. Researchers at Stanford University fed an AI model a library of two million bacteriophage genomes, asked it to invent new ones, and watched it generate thousands of candidates. Nearly 300 were built in a laboratory dish.
Only sixteen came alive — but those sixteen were enough to change everything. Bacteriophages are viruses with a very specific grudge: they attack bacteria and nothing else. Doctors in Georgia, Poland and Bangladesh have used them for decades to treat patients whose infections laughed off every antibiotic on the shelf. Now Dr Brian Hie, a chemical engineer at Stanford, has shown that an AI — using something called a genome language model, essentially the genetic cousin of the software behind ChatGPT — can design brand-new phages that never existed in nature.
A cocktail of his AI-built viruses killed strains of E. coli that were resistant even to natural bacteriophages. The bacteria had no answer for something evolution had never prepared them for. The excitement, however, is running alongside a chill.