Artificial intelligence has taken another step into synthetic biology, with researchers using a powerful AI model to design functional viruses from scratch in a breakthrough that could advance medicine while raising new biosecurity concerns.
In a study published Thursday, researchers at Stanford University and California’s Arc Institute said they used an AI model called Evo to engineer brand new viral genomes. The model was trained on trillions of genetic building blocks, teaching it the complex hidden “grammar” of natural DNA.
The AI generated hundreds of custom genetic recipes. When researchers chemically mixed those digital blueprints and introduced them into bacterial host cells, 16 of the AI-designed genomes produced fully functional viruses never before seen on Earth.
The synthetic viruses are not capable of infecting humans. Instead, they are bacteriophages based on the well-studied Phi X174 virus, meaning they infect only E. coli bacteria. In lab testing, the engineered phages successfully killed resistant bacterial strains that had evolved immunity to naturally occurring viruses.
A massive medical breakthrough may be at hand, as doctors could successfully fight antibiotic-resistant superbugs and deliver targeted gene therapies.
But some biosecurity experts are warning of dark possibilities.
While the Arc Institute researchers took strict safety measures, watchdogs warn the core technology opens the door to danger. If an AI can master whole-genome design to construct bacteria-killers, it could be only a matter of time before high-tech bad actors attempt to prompt similar language models to invent hyper-lethal, highly transmissible human plagues.
“You could say, ‘Hey, genomic language model, make me an influenza genome that is modified to be more transmissible or to be more lethal,’” Dr. Moritz Hanke, a fellow at the Johns Hopkins Center for Health Security, cautioned.
Adding fuel to the fire, biosecurity testers have already caught commercial AI mapping out high-level strategies for evading security, considering airborne dispersal methods, and generating tricky DNA variants that could slip past laboratory screening filters. In one alarming real-world case, authorities in India arrested a suspect who allegedly used AI chatbots while researching how to extract the deadly toxin ricin from castor beans.
Recent federal guidelines restricting high-risk laboratory research stop short of prohibiting purely computer-based bio-design, leaving AI-created DNA blueprints in a regulatory gray zone.
