In brief
U.S. researchers at Stanford used AI models Evo1 and Evo2 to design 16 entirely new bacteriophages that can replicate in the lab and kill E. coli. The work marks the first time a full viral genome has been generated by generative AI, opening a potential new frontier for treating antibiotic‑resistant infections. While hailed as a "significant turning point", the breakthrough also fuels urgent biosafety and biosecurity concerns, as experts warn that the same technology could be repurposed to create harmful pathogens. The team employed safeguards—excluding all human‑infective genomes from the training data and limiting work to phages—in order to minimise risk. The study signals a future where computers can write biology, but also underscores the need for robust policies to prevent misuse.
U.S. scientists at Stanford University have used generative AI to create 16 brand‑new bacteriophages that are fully functional in a laboratory setting. These viruses can replicate and kill the common bacteria –“Escherichia coli”—paving the way for next‑generation treatments of antibiotic‑resistant infections.
The First AI‑Generated Viral Genome
The breakthrough comes after two AI models, called Evo1 and Evo2, were trained on a massive genetic database that included viruses, bacteria, plants and humans. While the models traditionally learn language from text, the researchers taught them the language of life, allowing them to propose new DNA sequences that form a viable viral genome. Of the 302 candidate designs, 16 successfully infected and destroyed bacterial colonies in vitro.
Implications for Phage Therapy
Phage therapy—using viruses that specifically attack bacteria—has been investigated for decades, but practical deployment has been hampered by limited numbers of known phages. The AI approach could rapidly expand the phage toolbox, offering customized solutions against emerging resistant strains.
Safety and Security Concerns
While the study was praised as a “significant turning point” in synthetic biology, experts have highlighted urgent biosafety questions. The Johns Hopkins Center for Health Security argued the research “does not simply raise whether generative viral genome design can exist, but whether it can be used without enabling serious harm.” The team mitigated risk by excluding any genomes that could infect human or complex organisms and performed all work inside a highly secure laboratory.
Future Directions in Synthetic Biology
This work is a first step toward using AI to write genomes for more complex organisms. As noted by synthetic‑biology experts, the ability of AI models to learn evolutionary design principles means the next phase could see creation of living microbes, advanced enzymes, or even new therapeutic antibodies. The promise is huge, but it also underscores the need for global governance to ensure these powerful tools are employed responsibly.
