AI Designs Functional Bacteriophages With Novel Genomes

The BBC reports that the researchers used generative models called Evo1 and Evo2, which predict biological sequences in an approach analogous to language models predicting text. Those models were trained on genetic codes from viruses, bacteria, plants, and people, then refined for bacteriophage design.

According to the BBC, the team selected 302 generated designs for laboratory synthesis, and 16 proved effective at killing *E. coli*. Brian Hie, a Stanford assistant professor, told the BBC: "This is a next step in the complexity that's designable by generative AI, this is the first time generative AI has been used to design a complete genome, it's something that can replicate and have other functions inside cells... this was new territory for us."

The result differs from producing a known viral genome from a published sequence. The Times reports that synthetic viral genomes have long been used to study viruses, vaccines, and antiviral drugs, whereas the new experiments generated previously unobserved genomic sequences that still yielded viable viruses. Patrick Cai, a synthetic biologist at the University of Manchester who was not involved in the research, called it "an important milestone" in comments to the Times.

Bacteriophages are of interest as potential tools against bacterial infections, including strains resistant to conventional antibiotics. Nature's September 2025 coverage of the underlying preprint reported that AI-designed phages killed resistant strains of *E. coli*. The BBC similarly frames the work as a potential route toward new disease treatments.

At the same time, the BBC reports that experts have raised urgent safety and security concerns about AI-designed viruses. The current experiments involved bacteriophages with bacteria-specific hosts, not human pathogens. That constraint is central to interpreting the immediate risk: the reported viruses are not human-infecting agents.

For ML and computational-biology teams, the finding provides an unusually demanding form of experimental validation. A generated genome is not useful merely because it resembles training data; it must encode a coordinated biological system that can replicate in a host and produce infectious progeny. In comparable generative-biology programs, that requirement makes wet-lab screening, host-range testing, sequence provenance, and release controls as consequential as model quality metrics.

The work also sharpens a governance challenge common to advanced biological design systems. Models that learn constraints governing benign phages can increase scientific capability while raising questions about access controls, screening, and evaluation for more hazardous applications. The BBC's reporting places those questions alongside the medical potential rather than treating either outcome as established.