Published : 10 Aug 2026, 09:25 PM
Updated : 08 Sep 2026, 09:31 AM
Scientists have designed functioning viruses using artificial intelligence for the first time, in a breakthrough that could open new frontiers in medicine but has also triggered urgent warnings about biosafety and misuse, The Guardian reports.
The viruses are bacteriophages -- a type of virus that targets only bacteria -- and were designed using genome language models, the genetic equivalent of the AI systems behind chatbots.
In laboratory tests, a cocktail of the AI-generated viruses successfully killed strains of E coli resistant to naturally occurring bacteriophages.
The work was led by Dr Brian Hie, a chemical engineer at Stanford University, whose team used AI models called Evo1 and Evo2, trained on genetic data from two million bacteriophages.
The genetic code of viruses capable of infecting humans, animals or plants was deliberately excluded from the training data to limit risk.
Of the nearly 300 candidate genomes the team selected for laboratory testing, only 16 proved viable -- but a cocktail of those successfully overcame resistance in two different E coli strains. The results were published in the journal Science.
The researchers themselves acknowledged that the work raised "important biosafety, biocontainment and biosecurity considerations" and urged scientists designing whole genomes to involve safety and security professionals throughout.
Experts not involved in the research echoed those concerns. Writing in a commentary accompanying the study, Prof Tom Inglesby and Dr Moritz Hanke of Johns Hopkins University warned that while generative AI can now compose viral genomes, the governance framework to safely manage that capability does not yet exist.
Prof Tom Ellis of Imperial College London called the findings impressive, but cautioned that designing more complex viral genomes would be far harder.
He also argued that the threat of AI-designed pathogens was "very overblown" compared to the more immediate risk of scientists modifying existing dangerous viruses.
Dr Filippa Lentzos of King's College London said regulation should not focus solely on AI models.
A layered approach -- covering model access, research oversight, DNA synthesis screening and laboratory biosafety -- would be more effective, she said.
