AI creates 16 functioning viruses from scratch in genomic study

The breakthrough lies in what the AI learned from scale. By training on over 9 trillion nucleotides spanning 128,000 genetic sequences from millions of organisms, Evo discovered the underlying statistical patterns governing how DNA building blocks combine to form viable biological sequences. Before this, humans knew the individual nucleotides but lacked the formula for arranging them into functional genes—a gap that existed simply because the combinatorial space was too vast for human exploration. The study chose the Phi X-174 virus as a proof of concept because it is simple (only a few thousand nucleotides, compared to the three billion in humans) and extensively characterized, making it an ideal test subject.

The practical demonstration adds credibility to the findings. When researchers exposed E. coli strains that had evolved resistance to natural Phi X-174 to "cocktails" of the AI-designed phage variants, the AI-designed viruses evolved during the experiment and ultimately overcame bacterial resistance in all three strains. This shows the new organisms are not merely novelties but functionally competent—they can replicate, adapt, and exert biological effects just as natural viruses do.

The regulatory concern is substantial. The researchers themselves note that experts worry such studies are "way ahead of necessary guardrails and regulations." The risk is not the current work (focused on E. coli-only pathogens) but the pathway it opens: if the same method can generate novel viruses that infect bacteria, the same approach could—in principle—be applied to human pathogens, with consequences that the researchers explicitly flag as beyond the scope of their current safeguards.