First AI-Designed Viruses: Evo 2 Creates 16 Novel Phages That Beat Nature

Stanford researchers used the genomic AI model Evo 2 to design 16 novel bacteriophages — several of which infect and kill E. coli faster than their natural template — in what is described as the first AI-created viruses not found in nature, raising both therapeutic promise and biosecurity concerns.

Key Takeaways

Stanford researchers used the genomic AI model Evo 2 to design 16 novel bacteriophages — several of which infect and kill E.
coli faster than their natural template — in what is described as the first AI-created viruses not found in nature, raising both therapeutic promise and biosecurity concerns.

AI synthesis4 sources9

Researchers at Stanford University have used the genomic AI model Evo 2 to design and synthesize 16 entirely new bacteriophages — small viruses that infect bacteria — in what is being described as the first time AI has created viruses not found in nature. Several of the AI-designed phages infected and killed E. coli faster than their natural template, the ΦX174 phage.

The work, published in Science on August 6, was built on Evo 2, a genomic foundation model trained on genetic data from roughly 2 million bacteriophages. Rather than merely tweaking existing genomes, the model generated complete phage genomes from scratch. The team then synthesized the designs in the lab, confirmed they folded and functioned, and watched the viruses replicate and spread — some outperforming nature's own design.

Because the researchers deliberately excluded human-pathogen data from the training set, the phages are not dangerous to people. But the milestone has already ignited a debate about dual-use risk. Biosecurity experts, including Dr. Moritz Hanke, warn that the technique could lower the barrier to designing bioweapons, while Imperial College's Tom Ellis suggests built-in genetic restrictions could mitigate the danger — for instance, hard-coding dependence on synthetic amino acids that do not exist in nature.

The achievement is also a step toward a practical goal: engineered phages tailored to kill specific bacteria are seen as a potential answer to antibiotic resistance, and AI-designed genomes could dramatically accelerate phage therapy development.