Stanford and Arc Institute Create Viruses Designed Entirely by AI

For the first time, researchers have synthesized brand-new, self-replicating viruses based on genomes designed entirely by artificial intelligence. Scientists from Stanford University and the Arc Institute, a California nonprofit focused on high-risk, high-reward research, published the work in the journal Science on Thursday, August 6, as first reported by ABC News on August 7.

Genome Language Models as Designers

The team used two "genome language models" — Evo 1 and Evo 2 — trained on genetic code from viruses and more complex organisms, and instructed them to generate complete genomes for a viable bacteriophage: a virus that infects and replicates inside bacteria. The models produced DNA sequences that were then synthesized and introduced into E. coli cultures.

Viable, and in Some Cases Lethal, Phages

Shortly after inoculation, clear spots appeared in 16 of the Petri dishes where the bacteria were growing — evidence that the AI-designed viruses were attacking and replicating inside the E. coli. Some of the viable viruses proved more effective at attacking E. coli than the original ΦX174 bacteriophage the models were trained against, according to the researchers.

Implications and Guardrails

The demonstration suggests generative models can now explore biological sequence space far beyond what evolution has produced, opening the door to designed phages for medicine and agriculture — where bacteriophages are used as alternatives to antibiotics — as well as new tools for biotechnology. It also raises familiar dual-use questions about AI-generated biological agents, making the paper's safety discussions as closely watched as its results.