Researchers at Stanford University and the Arc Institute have achieved a major breakthrough by using artificial intelligence to design and create functional viruses in the laboratory. The AI model, named Evo, uses machine learning techniques similar to those behind large language models like ChatGPT, but instead of generating text, it generates genetic code.
Evo was trained on around two million genomes of bacteriophages, which are viruses that infect bacteria. Through this training, the AI learned the underlying patterns, structures, and organization of viral genes. It then applied that knowledge to design entirely new viral genomes, effectively “imagining” new, never-before-seen viruses based on what it had learned from existing ones. Out of 302 AI-generated genome designs, 16 were synthesized and tested in the lab using E. coli bacteria. Remarkably, these synthetic viruses were able to replicate and destroy bacterial cells, demonstrating that the AI’s creations were biologically functional.
This marks one of the first successful demonstrations of machine learning–based generative design in biology, where algorithms don’t just analyze data but actively create living systems. The implications are profound: the technology could accelerate drug development, genetic research, and bioengineering, making it possible to design vaccines, antibiotics, or gene therapies far faster than traditional methods allow.
However, the researchers also cautioned against the potential misuse of such technology. While Evo was trained only on harmless bacterial viruses, the same approach could, in theory, be applied to more dangerous pathogens, raising serious biosecurity and ethical concerns. Experts, including genomics pioneer J. Craig Venter, warned that without strict oversight, AI-assisted genome design could be exploited for harmful purposes.
Stanford’s study is demonstrating both the transformative power and potential dangers of combining AI with biology, a step that could redefine the boundaries of synthetic life and biotechnology