Researchers have used generative AI to create complete viral genomes that were later shown to work in the laboratory. The viruses were bacteriophages (phages), tiny viruses that infect bacteria, not humans.
The team generated 302 AI-designed phage genomes and tested the most promising candidates. Sixteen successfully produced functional viruses capable of attacking and killing E. coli.
That may sound like science fiction. It isn’t.
Meet the Researchers Behind the Breakthrough
The work involved researchers across Stanford University, the Arc Institute, and other institutions, with Brian Hie, an assistant professor of chemical engineering and data science at Stanford, playing a leading role. His research sits at the intersection of AI, genomics, and synthetic biology.
The team developed AI models known as Evo, including Evo 1 and Evo 2, designed to understand DNA in a way similar to how language models understand human language.
Think of ChatGPT predicting the next word.
Evo predicts what could come next in the genetic code.
Evo 2 was trained on an enormous collection of genomic data—about 9 trillion DNA base pairs—allowing it to recognize patterns across many forms of life.
So, What Exactly Is a Bacteriophage?
A bacteriophage is essentially a bacteria-hunting virus.
Unlike viruses that infect humans, phages target specific bacteria. They attach to a bacterial cell, enter it, multiply, and ultimately destroy the cell.
That makes them particularly interesting for medicine.
As antibiotic resistance grows, scientists are searching for new ways to kill dangerous bacteria. Phage therapy is one promising approach, and AI could potentially speed up the discovery of useful phages.
Instead of searching through nature for the right virus, scientists may eventually be able to design candidates on a computer and test the best ones in the lab.
Also Read | “What is the Chandipura virus?” Gujarat outbreak leaves 23 dead
Why This Is a Big Deal
Previously, scientists mainly used AI to analyze biological data, predict proteins, or improve existing molecules.
Now, AI is moving toward something much more powerful:
Designing biological systems from scratch.
The researchers’ results show that some AI-generated genetic instructions can actually be turned into functioning biological systems.
That is a major shift, from reading the code of life to writing new code.
The Biosecurity Question
If AI can design useful viruses, what happens when someone tries to design harmful ones?
That is why experts are calling for strong biosafety and biosecurity safeguards as these systems become more powerful.
The researchers took precautions, including focusing their work on bacteriophages rather than human-infecting viruses. Evo 2 itself was also designed with safety considerations, including excluding viral genomes from its training data.
The technology is nowhere near creating complex organisms or human diseases from scratch. But the direction is clear.
AI has learned to write biological code, and some of that code actually works.
Today, it’s bacteria-killing phages.
Tomorrow, it could be new medicines, new therapies, and entirely new ways of engineering biology.
The biggest question isn’t whether AI can design life.
It’s whether humanity can stay ahead of what AI learns to design next.