Researchers at Stanford University and the Arc Institute used AI to design 16 previously unknown bacteriophages—viruses that infect bacteria—which were shown to function in laboratory tests. The work could help develop alternatives to antibiotics against drug-resistant infections, but it also highlights concerns that similar tools could be misused to engineer harmful biological agents.

The team used Evo 1 and Evo 2, AI models trained on millions of genomes, to generate thousands of virus designs using the E. coli-infecting phage Phi X-174 only as a structural reference. From these, scientists selected and chemically synthesised 300 genomes; 16 produced functional viruses with novel sequences, regulatory elements and genome sizes. The study, published in Science, also tested AI-designed phages against E. coli strains resistant to Phi X-174, though the supplied article text does not include the experiment’s full outcome.