Stanford-Arc Institute team uses AI to design 16 viable synthetic viruses not found in nature

AI generates functional viral genomes

Researchers at Stanford University and the Arc Institute in Palo Alto used genome language models called Evo 1 and Evo 2 to design synthetic bacteriophages, viruses that infect only bacteria. The models were trained on trillions of nucleotides and further on approximately 15,000 viruses from the same family as Phi X-174, a tiny virus that exclusively infects E. coli. The Guardian reports the models were also trained on genetic data from 2 million bacteriophages. Genetic code for viruses capable of infecting plants, humans or other animals was intentionally excluded from training to reduce the risk of designing dangerous pathogens. The work was published on Thursday, 6 August 2026, in the journal Science.

The AI generated roughly 700,000 potential genomes, from which the team selected 285 candidates (described as "about 300" or "nearly 300" in other reports) for laboratory synthesis. The DNA molecules were inserted into bacteria, which read the genetic code and produced the new viruses. Only 16 proved viable, but those 16 were at least as viable as the natural Phi X-174, with some reproducing faster. A cocktail of the AI-designed phages overcame resistance in two different strains of E. coli, suggesting they were more than crude imitations of naturally occurring viruses.

Medical potential

Brian Hie, assistant professor at Stanford and the project leader, described the work as a first demonstration that generative AI can produce entire functional genomes. The researchers chose phages because they are small, well-studied, and have broad applications in molecular biology and therapeutics. Phages are sometimes used instead of antibiotics to kill bacteria causing disease, and AI could help revive phage therapy as antibiotic resistance grows.

Our study is a proof of concept showing for the first time that generative design can generate entire functional genomes.

Samuel King, another member of the Stanford team, said that emerging antibiotic-resistant pathogens will require innovative alternative solutions, and that phage therapy is one such option. Adrian Woolfson, a genomics expert and chief executive of the DNA synthesis company Genyro, called the work biology's Wright Brothers moment.

It is biology's Wright Brothers moment, where we stop looking over our shoulder at the entities that evolution has created... and are confronted with the vast expanses of future biological possibility.

Patrick Cai, a synthetic biologist at the University of Manchester who was not involved in the study, called it "an important milestone." Hie added that the approach could be extended to larger organisms such as bacteria and then to more complex forms of life.

Biosafety and governance concerns

Tom Inglesby and Moritz Hanke of the Center for Health Security at Johns Hopkins University wrote an accompanying article in Science praising the Stanford team for taking precautions but warning that existing guardrails are insufficient.

Although this is promising for life sciences applications, it also raises urgent biosafety and biosecurity questions. The ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not.

The researchers themselves urged others designing whole genomes to consult both safety and security professionals throughout their projects. The work also intersects with US regulatory efforts. The Trump administration issued a policy in July 2026 prohibiting federally funded "gain of function" research and calling for enhanced oversight of projects involving harmful biological agents, but that policy did not specifically address AI-driven research. On the same day the study was published, Senator Rand Paul's committee held Anthony Fauci in contempt for refusing to answer questions from lawmakers investigating the origins of the COVID-19 pandemic. The AI-designed viruses, however, represent something entirely new rather than manipulation of natural pathogens, placing them outside existing regulatory frameworks designed for traditional gain-of-function research. Biotechnology in the US is regulated through a patchwork of laws and agencies, and breakthroughs frequently outpace regulators.