Phage against the machine: AI-designed viruses stick it to E. coli

For the first time, scientists have used AI to design viruses that kill bacteria. This advance has major medical promise, but could pose serious safety and biosecurity questions.

Viruses: Nature’s hackers

Viruses are genetic instructions in a protein shell. Most experts don’t classify them as alive but anyone who’s sweated through a fever knows they sure can reproduce.

They do this by hijacking cells; binding, injecting their code, and turning the host into a factory. Once packed with thousands, even millions, of these micro-marauders, the cell bursts, seeding nearby tissue and fluids.

But some viruses are useful. Bacteriophages, for example, target bacteria.

And now we have enlisted AI to help us create helpful viruses.

Byte me

As reported in Nature on 18 September 2025, researchers trained models (Evo1/Evo2) on over 2 million phage genomes to generate synthetic designs resembling a well-understood virus ΦX174.

I should point out this AI is not a large language model, the platforms that have been all the rage for the last three years. Evo1/Evo2 is a generative neural network trained on DNA sequences. Closer to the systems that broke chess and Go a decade ago than to ChatGPT.

Flick that kill switch

And as if the name ΦX174 isn’t cool enough by itself, these lab-made variations were tweaked to combat three strains of drug-resistant E. coli that the natural virus could not infect.

They synthesized 302 AI-designed candidate genomes. Of these 302 potential killer-codes, 16 produced live, infective phages that killed resistant bacteria in lab tests. Even better, combinations of the phages conquered all three strains.

This is the first time fully AI-designed viral genomes have been shown to kill drug-resistant bacteria.

In your face, E. coli!

As the researchers Samuel King and Brian Hie themselves noted in Arc Systems:

“The transition from reading and writing genomes to designing them represents a new chapter in our ability to engineer biology at its foundational level.”

To V or not to V?

The study is a preprint, and safety questions loom. Could these viruses mutate, harm beneficial microbes, or be misused with malicious intent?

As genetics legend J. Craig Venter, the scientist celebrated for sequencing the human genome, told Newsweek:

“I would have grave concerns if similar random viral enhancement approaches were applied to smallpox or anthrax.”

The promise is real, but so are the biosecurity risks.

If handled carefully, this could be a paradigm shift in medicine. If not, we may learn the hard way what it means to play coder with evolution.

Adam S