AI Designs First Working Viruses to Kill Bacteria

The AI tools, called Evo1 and Evo2, work like the large language models behind AI chatbots, but they predict genetic code instead of words. Reports about their training differ. The Guardian said the models learned from two million bacteriophages, and that code from viruses which infect humans, animals or plants was left out on purpose. The BBC (the British Broadcasting Corporation) said the training was wider and was narrowed to phages later. Hie stressed that all the work happened in a secure laboratory. Doctors in several parts of the world already use phages against stubborn infections, and interest is growing as bacteria become resistant to antibiotics.

Experts called the result important but limited. Marc Güell, a scientist in Spain, described it as a "very significant turning point". Tom Ellis of Imperial College London noted that a phage genome has only about 5,400 base pairs, while the simplest living cell has around 500,000. Writing in Science, Tom Inglesby and Moritz Hanke of Johns Hopkins University warned that the power to build viral genomes now exists, but the rules to control it do not. Nobody, they said, should design viruses able to infect people, animals or plants.