AI-Designed Viruses Kill E. Coli—and Expose a Governance Gap
Researchers created 16 functional bacteriophages from AI-written genomes, opening a possible route to new treatments for resistant infections while exposing gaps in biosecurity oversight.
This A.I. Just Created Viruses Not Found in Nature
AI used to create viruses not found in nature for first time
This A.I. just created viruses not found in nature
Scientists have created new viruses using AI. Here’s how they did it
Scientists use AI to create viruses that don’t exist in nature for the first time
A laboratory milestone with two edges
Researchers at Stanford University and the Arc Institute have used generative AI to write complete genomes for new bacteriophages, viruses that infect bacteria. After synthesising roughly 300 promising designs, the team found 16 that could reproduce and kill E. coli in laboratory dishes.aljazeera +1 The result, published Thursday in Science, is the first demonstration that an AI-generated genome can produce a functional virus not found in nature.nytimes +1
The experiment does not mean AI has created a human pathogen. The phages were modelled on Phi X-174, a tiny virus that attacks bacteria, and the viable designs posed no threat to people.nytimes +1 Their significance lies instead in proving that a genome language model can move from predicting genetic sequences to specifying a biological entity that replicates.
From genetic language to working phages
The team used the Evo 1 and Evo 2 models, which predict sequences of DNA much as a large language model predicts text. Researchers excluded viruses capable of infecting humans, animals or plants from the relevant training data, then asked the models to generate complete phage genomes and inserted selected synthetic DNA into bacteria.theguardian +1 Only 16 candidates worked, underscoring how difficult whole-genome design remains.theguardian
The successful phages were more than a technical proof. A cocktail of them overcame resistance in two strains of E. coli, and some performed better than the natural phage used as a template.theguardian +1 That points toward faster development of customised phage therapies for infections that no longer respond to antibiotics, although laboratory success is far from a treatment ready for patients.aljazeera +1
Governance trails the capability
The same method creates a dual-use problem. Johns Hopkins health-security experts Thomas Inglesby and Moritz Hanke warned that the ability to compose viral genomes now exists while governance capable of steering it safely does not.axios +1 The Stanford team used a secure laboratory and deliberately limited the organisms in scope, but those precautions are not universal requirements.theguardian +1
There are still formidable barriers to misuse. Phi X-174 has only about 5,400 DNA base pairs, while the smallest cellular genomes are around 500,000; synthesising and testing designs also requires substantial wet-lab capability.aljazeera +1 Even so, experts are calling for layered safeguards spanning model access, research review, DNA-synthesis screening and laboratory security.theguardian The immediate achievement is a potential new tool against resistant bacteria. The larger consequence is that oversight must now address AI systems that can write functioning biology, not merely analyse it.