aceAI-designed bacteriophage genomes
Source article: Safety fears as scientists make first viruses designed by AI
Abstract: Scientists have made the first viruses designed by artificial intelligence in a milestone that raises hopes for new medicines but also concerns over how to ensure the technology remains safe. The viruses are specific kinds known as bacteriophages, which only infect bacteria and are used around the world to treat patients with persistent infections. In lab tests, a cocktail of the AI-designed viruses killed E coli bugs that were resistant to natural bacteriophages. Dr Brian Hie, a chemical engineer at Stanford Un…
Negative state: both sides are scored from claims and sources, not community votes.
On 2026-08-06, researchers led by Stanford reported the first viruses whose genomes were designed by AI. Using models called Evo1 and Evo2 trained on 2 million bacteriophage sequences, they generated thousands of designs, built nearly 300 in the lab, and obtained 16 viable bacteriophages that as a cocktail killed E. coli resistant to natural phages in dish experiments.
The result matters because it demonstrates generative AI can create functioning viral genomes that could be tuned to overcome resistance and expand phage therapy for persistent infections, but it also proves the capability exists before governance does. Uncertainty remains about efficiency, scalability beyond tiny phage genomes, and how to prevent misuse for more dangerous viruses.
- Stanford researchers used genome language models Evo1 and Evo2, described as the genetic equivalent of large language models behind AI chatbots, to design bacteriophage genomes.
- Models were trained on genetic data from 2 million bacteriophages with code for viruses that infect plants, humans or other animals intentionally excluded.
- From thousands of AI-generated genomes, nearly 300 were synthesized and dropped into bacteria to produce phages; only 16 proved viable.
- A cocktail of the viable AI-designed phages overcame resistance in two different strains of E. coli in dish experiments.
Researchers used genome language models Evo1 and Evo2 trained on 2 million bacteriophage genomes to design functional bacteriophage genomes, and a cocktail of the resulting viruses killed E. coli strains resistant to natural phages in lab dishes.
The same ability to compose functioning viral genomes with generative AI raises urgent biosafety, biocontainment and biosecurity considerations, with commentators warning governance to safely steer the technology does not yet exist.
The rundown
The team generated thousands of candidate genomes and selected nearly 300 to synthesize, using bacteria as factories that read the code and churned out new bacteriophages for testing against E. coli.
The work was published in the journal Science on 2026-08-06, with an accompanying commentary from Johns Hopkins Center for Health Security urging consultation with safety and security professionals and caution against pursuing human, animal or plant pathogen genomes.
The approach was inefficient and limited to the simplest viral genomes, with only a small fraction of designs viable and applicability to more complex or human-infecting viruses unknown.
Sources
- JournalismThe Guardian2026-08-06
How should this claim be treated?
The debate