Stanford Researchers Unveil AI-Powered Laboratory for Accelerated Scientific Discovery

Researchers at Stanford University reported that they used generative artificial intelligence to create fully synthetic viruses that can replicate under laboratory conditions. The work, published in Science , describes 16 designed viruses that were built from AI-generated genome sequences and then shown to function in controlled experiments.

Officials involved in the study said the project is the first time AI models have produced complete viral genomes that are both novel and viable, moving beyond partial sequence suggestions. The researchers targeted bacteriophages, which infect bacteria rather than human cells, keeping the experiments focused on non-human pathogens.

How the Evo1 and Evo2 models generated viral genomes The research team used two models, Evo1 and Evo2, described as operating in a way similar to large language models. Instead of predicting the next word in a sentence, the systems predict genetic sequences, using patterns learned from diverse biological data.

According to the study description, the AI produced 302 candidate viral genome designs. Researchers then selected designs for physical synthesis and laboratory testing, bridging a key gap between digital sequence generation and real-world biology.

From hundreds of candidates to a smaller set of viable viruses Out of the 302 AI-generated candidates, 16 viruses were found to be fully functional in the lab, meaning they could replicate in that environment. In petri dish experiments, those 16 also demonstrated an ability to eliminate E. coli bacteria under controlled conditions, the researchers said.

The reported workflow highlights an end-to-end pipeline: computational design, synthesis of genetic material, and lab validation. The researchers framed this as a notable step for synthetic biology because it translates probabilistic sequence prediction into organisms that perform measurable biological functions.

Medical promise and biosafety scrutiny

The study’s focus on bacteriophages connects to ongoing interest in alternatives to conventional antibiotics. The researchers said the approach could point toward future options for addressing antibiotic-resistant infections, though the work described here remains in controlled laboratory testing and is limited to bacteria-infecting viruses.

At the same time, the capability drew immediate attention from biosafety and security experts. Specialists from the Johns Hopkins Center for Health Security said that the demonstrated ability to design new viral genomes increases the urgency of creating and updating security frameworks aimed at preventing misuse.

Governance questions raised by synthetic genome design

While the current experiments were confined to bacteriophages and not directed at human pathogens, the methodology shows that AI systems can help generate biological entities that do not exist in nature, based on the researchers’ description. That shift, experts said, makes long-term governance of synthetic genome design a central issue alongside scientific progress.

The researchers and external experts pointed to a shared uncertainty: how to build oversight that supports legitimate research while reducing the risk of harmful applications as genome-design capabilities become more accessible.