Credit: sanjeri / Canva
Researchers at Stanford University and the Arc Institute reported the milestone on Thursday in the journal Science, in work detailed by The New York Times. Using AI trained on genetic code rather than text, they generated the genomes of 16 new viruses. Each is a bacteriophage, a virus that infects only bacteria.
The technique borrows straight from chatbots. Models called Evo 1 and Evo 2 work like ChatGPT for biology. They were trained on millions of genomes to learn the “grammar” of DNA. Prompted with a fragment of a natural virus, they wrote out full genetic blueprints for new ones.
How they did it, and what they got
The team used a tiny, well-studied virus called ΦX174 as a template. Evo generated roughly 700,000 candidate genomes. The scientists narrowed these to about 300, built them in the lab, and dropped them into E. coli. Just 16 produced working viruses. It is a low hit rate, but the survivors were real.
They were not mere copies, either. Some carried new genes, or swapped in parts from distantly related viruses. Most striking, a cocktail of the AI viruses quickly overcame E. coli strains that had grown resistant to natural ones. That points to a genuine prize: better therapies against bacteria that antibiotics can no longer kill.
The part that worries people
The researchers built in guardrails. They excluded viruses that infect humans, animals, or plants from the training data, worked only on harmless phages, and ran everything in a secure lab. But the method is now public, and it will not stay confined to bacteria forever.
Writing alongside the study, Johns Hopkins biosecurity experts Tom Inglesby and Moritz Hanke put it starkly. “The ability to compose viral genomes using generative AI now exists,” they wrote, “the governance to safely steer it does not.” Chatbots can already be coaxed toward biological weapons; a tool that writes genomes raises the stakes.
Not everyone is alarmed. Tom Ellis of Imperial College London called ΦX174 “literally the smallest and easiest genome to make,” and told The Guardian the threat is “very overblown” next to simply altering existing pathogens. Others see a turning point: one DNA-synthesis chief called it biology’s “Wright brothers moment.”
Rules that have not caught up
The policy gap is real. Last month the US administration issued a policy to curb high-risk life-sciences work, but it targets “gain of function” experiments on natural pathogens, not purely computational design. AI that dreams up new genomes on a screen falls through the gap, even as Washington writes voluntary rules elsewhere.
The fair reading is that this is a real scientific leap, with real upside, achieved by a team that flagged its own risks. It is also a preview. Experts in AI biosecurity argue the place to intervene may be where DNA is physically manufactured, not just the model. The science has arrived. The rules have not.
Get the TNW newsletter
Get the most important tech news in your inbox each week.
