Researchers have successfully used artificial intelligence (AI) to design entirely new viruses not found in nature, marking a significant development with potential applications in medicine and biotechnology, alongside raising biosecurity concerns. The study, conducted by scientists at the Arc Institute in Palo Alto, California, and Stanford University, was published on August 6 in the journal Science.
The research focused on bacteriophages—viruses that infect bacteria but do not harm human or animal cells. Specifically, the team concentrated on Phi X-174, a well-studied bacteriophage that infects Escherichia coli (E. coli) bacteria, chosen for its relatively small genome and its safety profile. Using an AI model called Evo, a genome language model trained on genetic sequences from millions of organisms including bacteriophages, the researchers generated approximately 700,000 potential viral gene sequences based on patterns learned across the tree of life.
From these, scientists synthesized DNA for 285 novel genomes and tested their viability by inserting them into bacteria. In total, 16 AI-designed viruses successfully infected and replicated within bacterial cultures. Some even demonstrated faster replication rates than their naturally occurring counterparts. The results suggest that AI can generate functional viral genomes that may serve as useful tools for gene therapy, biotechnology, and studying viral biology.
The AI models, Evo 1 and Evo 2, operated by recognizing the "grammar" of DNA sequences similar to how language models process text, enabling the creation of genomes from scratch rather than copying existing ones. Researchers emphasized they excluded genetic data from viruses capable of infecting humans, animals, or plants to minimize risks, focusing only on bacteriophages for safety reasons.
Despite these precautions, the study has elicited warnings from biosecurity experts who highlight a gap in current oversight frameworks. While the U.S. National Institutes of Health (NIH) recently introduced policies aiming to restrict high-risk biological experiments, these measures do not fully address computer-based activities such as AI-generated DNA that may potentially create novel pathogens. Experts note the difficulty in assessing threats posed by entirely new organisms without natural precedents, underscoring the risk of accidental or malicious misuse of AI in biology.
Dr. Moritz Hanke, from the Johns Hopkins Center for Health Security, noted that governments and scientific bodies have been slow to establish comprehensive safeguards for rapidly advancing technologies like AI-driven genome design. While acknowledging the medical promise of AI-designed viruses, he and others stress the urgent need for regulatory guardrails to prevent misuse that could facilitate the creation of dangerous biological agents.
The researchers behind the study maintain that their work was conducted with caution, deliberately limiting AI training data to exclude harmful viruses. Nonetheless, their breakthrough highlights how AI's growing capabilities in synthetic biology could reshape disease research and therapy, while simultaneously prompting ongoing debates over biosafety and biosecurity protocols in an emerging scientific frontier.