AI Models Design Functional Synthetic Viruses in Biological Engineering Breakthrough

Researchers at Stanford University and the Arc Institute have used generative artificial intelligence to design functional synthetic viruses capable of infecting and reproducing inside host bacteria, according to a landmark study published in the journal Science.

The development marks a major milestone in synthetic biology, proving that biological foundation models trained on raw genetic code can assemble working genetic architectures from scratch. The technology opens new avenues for gene therapy and targeted antibacterial treatments while sharpening scrutiny around biological security guardrails.

Out of roughly 700,000 potential viral genomes generated by the AI models—known as Evo 1 and Evo 2—scientists synthesized DNA for 285 selected candidates and introduced them into host cells. Laboratory tests revealed that 16 of the AI-designed candidates produced fully viable viruses capable of successfully infecting and replicating inside Escherichia coli bacteria, with several reproducing at speeds matching or exceeding natural strains.

The computational models operate on principles similar to large language models like ChatGPT, but instead of analyzing textual syntax, they process biological sequences across trillions of nucleotide base pairs to master the regulatory logic of DNA. To direct the experiment, researchers refined the models using approximately 15,000 viral genomes from the family of Phi X-174, a historic bacteriophage that in 1977 became the first DNA-based organism ever fully sequenced.

Scientists explicitly restricted the system’s design parameters to bacteria-targeting viruses, ruling out sequences capable of infecting human, animal, plant, or fungal organisms. The research team highlighted that customized bacteriophages could accelerate precision medicine, enabling tailored therapies to attack antibiotic-resistant bacterial strains and deliver therapeutic genes directly to host cells.

However, the capability of generative AI to engineer functional life forms has amplified biosecurity concerns regarding potential dual-use applications. While commercial gene-synthesis providers maintain biosecurity screening protocols to prevent the production of dangerous pathogens, the study demonstrates that molecular design barriers are falling rapidly as artificial intelligence moves from predicting biological structures to generating novel living entities.