Stanford and Arc Institute AI Designs 16 Novel Viruses From Scratch

Researchers from Stanford University and the Arc Institute have used a genomic language model to design 16 viable, functional viruses that do not exist in nature. The breakthrough demonstrates that AI can now autonomously draft biological blueprints for living entities, moving beyond the modification of existing organisms to the invention of entirely new ones.

To achieve this, the team utilized a model named Evo, which was trained on a massive dataset of approximately nine trillion nucleotides from animals, plants, microbes, and viruses. The study specifically targeted bacteriophages—viruses that infect bacteria—by training the AI on the Phi X-174 virus, which infects E. coli, along with roughly 15,000 related sequences. From an initial pool of 700,000 potential variations generated by the AI, the researchers synthesized 285 sequences, which ultimately yielded 16 fully viable novel viruses.

A Shift in Biological Design

While the laboratory synthesis of viral genomes is a standard scientific practice, this research represents a fundamental shift in methodology. Traditionally, scientists have edited or tweaked known viral strains to study their behavior. In this instance, the researchers used AI pattern recognition to learn the "grammar" of genetic sequences, allowing the model to invent biological structures from scratch. To mitigate the risk of creating dangerous pathogens, the team deliberately excluded human-infecting viruses from the training data.

Biosecurity and Dual-Use Risks

The ability to algorithmically design viable pathogens has raised severe biosecurity and "dual-use" concerns among experts. The primary fear is that the same technology used for scientific discovery could be repurposed to engineer lethal agents. Moritz Hanke, a fellow at the Johns Hopkins Center for Health Security, noted that a genomic language model could potentially be asked to create an influenza genome modified to be more lethal or transmissible.

Oliver Crook, a protein chemist at the University of Oxford, emphasized the novelty of these creations, stating that they are not just "sickly versions of stuff that already exists." This capability suggests that AI can create robust, functional biological entities that bypass the limitations of natural evolution.

Regulatory Loopholes

The study highlights critical gaps in current oversight. Experts warn that existing regulatory frameworks, including policies from the National Institutes of Health (NIH), contain loopholes because they primarily prohibit the synthesis of known biological entities of concern. Because these AI-generated viruses are entirely novel and do not match any known sequence in a database, there is currently no mechanism to preemptively block the creation of AI-generated lethal agents. The scientific community now faces the challenge of updating safety protocols to address a world where pathogens can be designed by an algorithm.