Scientists use AI to create entirely new viruses for the first time

Scientists Use AI to Design Entirely New Viruses

Researchers at Stanford University and the Arc Institute have used artificial intelligence to design and synthesize viruses that do not exist in nature—a first that marks a potential inflection point in how biotechnology is developed, while raising acute biosecurity questions.

The team trained an AI model to recognize patterns in naturally occurring bacteriophage DNA, then used it to generate genetic sequences for brand new viruses. After synthesizing those sequences and inserting them into bacteria, the researchers found the bacteria produced viable viruses capable of infecting other bacteria. The work was published Thursday in Science.

The viruses pose no threat to humans. The AI was trained exclusively on bacteriophages—viruses that infect only bacteria—and specifically excluded viruses that infect humans, plants, animals, and fungi. The model cannot generate human-infecting pathogens as a result. Even these newly designed viruses target bacteria alone.

The material difference from prior work lies in speed and cost. Scientists have been synthesizing viral genomes in laboratories for years to study disease and develop vaccines. What's new is the use of AI to design functional viruses from scratch at scale. If the cost of generating viruses drops by orders of magnitude, biotech development could accelerate significantly. That same acceleration creates biosecurity risk: if the approach scales across other viral classes, the barrier to entry for creating dangerous pathogens could collapse. The nightmare scenario, as one speaker frames it, is a gaming PC with stock graphics cards running an open-source model that walks someone through designing a problematic virus.

The research highlights how quickly AI capabilities are outpacing regulatory infrastructure. Governments and scientific organizations have been slow to develop guardrails that could block the creation of deadly viruses even as the underlying science advances. The researchers are not expected to open-source this work. Other AI-driven biotech labs—including Jeff Dean's Discovery Loop, which focuses broadly on biotechnology, and narrower projects targeting specific diseases like the common cold—will likely face similar tension between capability and control.

One practical upside: viruses are already used as delivery vehicles for gene therapies and other medical treatments. AI-designed viruses could eventually expand that toolkit. But marketing may matter. The word "virus" carries overwhelming negative connotation post-pandemic. If biotech companies want to commercialize viral delivery systems at scale, they may need to rebrand the concept entirely—much as "GLP-1 peptide" felt safer and more welcoming than steroids or performance-enhancing drugs.