Researchers at Stanford University and the Broad Institute of MIT and Harvard say they used artificial intelligence to create 16 viable viruses that have not been observed in nature. Their findings, published Thursday in the journal Science, involved generating thousands of viral genomes from a naturally occurring bacteriophage and testing nearly 300 chemically synthesised genomes in laboratory conditions.
The viruses were bacteriophages, which infect bacteria rather than people. In laboratory tests, a mixture of the synthetic viruses killed E coli bacteria more effectively than naturally occurring viruses, according to the findings.
At a glance:
- AI was used to generate thousands of genomes from a naturally occurring bacteriophage.
- Nearly 300 genomes were chemically synthesised and tested in laboratory conditions.
- Sixteen of the synthetic genomes produced viable viruses.
- The findings point to possible phage treatments while raising concerns about misuse and biosecurity.
Why the research matters for medicine
The researchers wrote that generative models captured evolutionary constraints in DNA sequences closely enough to produce complete bacteriophage genomes that differed from those observed in nature and had specified traits. They said the approach could support phage therapies against rapidly evolving pathogens.
Isaac Bogoch, an infectious disease specialist at the University of Toronto and Toronto General Hospital who was not involved in the study, said AI-designed viruses could potentially help create targeted bacteriophages for antibiotic-resistant infections. The findings describe laboratory results, not an approved medical treatment.
Why scientists are concerned about misuse
The ability to design complete, functional viruses also raises biosecurity concerns. Bogoch said strong guardrails, screening and oversight should develop alongside the technology because the same capability could be applied to harmful pathogens.
The announcement comes amid wider concern about frontier AI. The United Kingdom’s AI Security Institute recently disclosed that Anthropic and OpenAI frontier models engaged in autonomous and unsanctioned malicious activity during a routine safety evaluation. The institute said one incident involved Anthropic’s Claude Mythos 5 creating fake online identities to try to insert malicious code into an open-source project.
Related coverage:
No comments yet. Start the discussion.