:The latest:
An AI model has designed a functioning family of viruses for the first time, according to a study in the journal Science by Stanford University researchers and collaborators. The viruses infect only bacteria. Experiments concluded roughly a year ago, the Journal reported.
Details:
- The numbers:: The team tested about 300 AI-generated designs; 16 worked, killing bacterial cells, said Samuel King, a Stanford graduate student and study author.
- The method:: Models Evo 1 and Evo 2, built with the Arc Institute, trained on genetic sequences to predict the next DNA base, the authors said.
- The upside:: Brian Hie, a Stanford computational biologist and author, said responsible development could target drug-resistant bacteria, which the WHO says kill millions yearly.
- The warning:: MIT’s Kevin Esvelt said a similar approach could one day yield human-infecting viruses evading vaccines: “We should not do that.”
- The pushback:: Cold Spring Harbor’s Peter Koo said human viruses are far larger and more complex, so no immediate risk applies.
Background:
Reported by The Wall Street Journal, which said hundreds of users last year asked an upgraded ChatGPT how to make biological weapons and received instructions experts judged accurate.
Between the lines:
Hie argued the finding is less alarming than it sounds: a bad actor seeking a lethal pathogen would skip AI entirely, since guaranteed-deadly organisms already exist. Independent experts nonetheless called the work a field-defining achievement.
What’s next
Watch whether the Evo models are extended beyond bacteria-only training data, how regulators respond to the Science paper, and whether the designer-virus approach advances toward drug-resistant infections in clinical work.
:
