Researchers Use AI Genome Models to Design 16 Viable Bacteriophages, Study in Science Reports
The Facts
- A study published Thursday in the journal Science described the use of AI to design complete, functional viral genomes, which researchers say is the first time whole genomes have been designed by AI.
- The work was conducted by researchers affiliated with Stanford University, the Arc Institute in Palo Alto, California, and the Broad Institute of MIT and Harvard.
- The team used genome language models called Evo 1 and Evo 2 — described as the genetic equivalent of the large language models behind AI chatbots — trained on DNA sequences, using the natural phage ΦX174 as a template.
- The models generated thousands of candidate genomes; approximately 300 were chemically synthesized and tested in the laboratory, and 16 produced viable viruses.
- The resulting viruses are bacteriophages that infect only bacteria — specifically Escherichia coli — and researchers say they pose no threat to humans.
- In laboratory tests, a combination of the AI-designed phages killed E. coli strains that were resistant to the natural bacteriophage they were derived from.
- Researchers and outside commentators say the potential medical application is phage therapy — custom-designed viruses to treat bacterial infections that no longer respond to antibiotics.
- The researchers say they deliberately restricted the work to viruses that infect bacteria and did not train the model on viruses that infect humans.
- Scientists and biosecurity specialists have said the results raise safety and security questions, including whether similar tools could eventually be used to design dangerous pathogens, and some have called for new regulatory rules; existing sources indicate no approval regime governed the design step itself.
Context
What is a bacteriophage, and why did researchers choose one?
Bacteriophages, or phages, are viruses that infect bacteria rather than human or animal cells, and they are already used in some countries to treat persistent bacterial infections Guardian. The team worked from ΦX174, a small, well-characterized natural phage, and targeted the bacterium Escherichia coli C Izvestia.ru,Life.ru. Because phages cannot infect people, the researchers described the choice as a deliberate limit on the risk of the experiment India Today,News18.
How is this different from synthesizing viruses, which scientists have done for years?
Researchers have long been able to manufacture viral genomes by copying known sequences, work used to study antiviral drugs and vaccines NYT. In this case the AI model was trained to recognize patterns across large amounts of genetic data and then generated novel genome sequences that do not exist in nature, which were subsequently synthesized as physical DNA and shown to produce replicating viruses NYT,Olhar Digital - O f…,infobae.
What are the stated safety concerns, and what is being proposed?
Experts quoted in coverage of the study described the biosecurity questions as urgent, noting that generative AI can now assemble an entire functioning viral genome BBC,Guardian,News18. In an accompanying commentary in Science, Thomas Inglesby and Moritz Hanke of the Johns Hopkins Center for Health Security argued the work shows AI is capable of designing dangerous biological agents and called for strict legal limits on future research Forbes. Reporting also notes that DNA synthesis remains a costly step routed through commercial suppliers, which is where screening controls currently sit infobae.
Where Left and Right agree, and where they split
- Where Left and Right agree
- The researchers' self-imposed limits — bacteria-only phages, no human-infecting viruses in training — were the right line, and the payoff against resistant E. coli is genuine.
- Where Left and Right split
- Whether the story is about a safeguard that only holds if it becomes binding on everyone, or about scientists who drew the line themselves and earned room to work.
How left and right read it
Researchers at Stanford, the Arc Institute and the Broad Institute used genome language models to design complete viral genomes, and 16 became viable bacteriophages that killed E. coli strains resistant to their natural counterpart. The promise for infections antibiotics no longer touch is real. So is the exposure. The team chose to work only on bacterial viruses and withheld human-infecting ones from training — a safeguard that should be public policy, not a courtesy.
“For the first time, scientists have used artificial intelligence to create new kinds of viruses, raising hopes for medical advances while also raising the disturbing possibility that the technology could someday be used to invent dangerous pathogens.” — The New York Times
The restraint came from the researchers themselves: phages that infect only bacteria, and a model deliberately not trained on viruses that infect humans. Discipline chosen at the bench beats discipline imposed from outside, and the result is real — designed phages killed E. coli resistant to the natural virus. As The Daily Wire put it, 'a massive medical breakthrough may be at hand.' Credit that self-drawn line, and leave the work room.
“A massive medical breakthrough may be at hand, as doctors could successfully fight antibiotic-resistant superbugs and deliver targeted gene therapies.” — The Daily Wire
The receipts — all 100 sources
Independent coverage (50)
Facts first. Then every angle.
The day’s biggest stories in one short brief — the facts everyone agrees on, then the competing values behind the headlines. Free in your inbox.