Scientists Use AI to Create First-ever Functional Synthetic Life, Because What Could Go Wrong?

AI-generated image.

While the world was busy arguing over AI-generated art and essays, a team at the Arc Institute and Stanford University was training AI on a much more complex language: the code of life itself. In a breakthrough study published on the preprint server bioRxiv scientists revealed they have successfully used generative AI to design viable bacteriophages — viruses that hunt and kill bacteria.

Traditional genetic engineering is a painstaking grind of trial and error. We’re often limited by our own “human” understanding of how thousands of genetic parts interact. To bypass this bottleneck, researchers turned to Evo, a new class of “genome language models.”

Just as ChatGPT predicts the next word in a sentence, Evo was trained on millions of pieces of genetic information to predict the next nucleotide in a DNA sequence. Evo 1 used ~2.7M genomes, Evo 2 used a curated atlas of 128k organisms (with 9.3 trillion nucleotides). By learning the “grammar” of viral evolution, the AI can begin to innovate, proposing holistic genomic architectures that have never existed in the wild.

The researchers used a well-known virus called ΦX174 as their template. It’s a tiny, elegant predator that infects E. coli bacteria. This virus is called a bacteriophage, meaning bacteria-eater. By “prompting” Evo with a few snippets of ΦX174 DNA, they generated thousands of synthetic candidates. After a rigorous digital filtering process, they took 285 of the best designs and physically synthesized them in the lab.

16 New Versions of a Virus

The results were striking. Out of those designs, 16 came to life. These weren’t clones or slight variations of the existing virus; they contained hundreds of mutations and entirely new structural solutions that have never been seen in nature.

One AI-designed phage, Evo-Φ36, even managed to use a DNA-packaging protein from a distantly related virus — a feat that previous “human” engineering attempts had failed to achieve. Some phages, like Evo-Φ2147, were so different from known viruses (<95% sequence identity) that they would technically qualify as entirely new species. Another variant, Evo-Φ69, dominated the population in every trial, replicating and spreading significantly faster than the wild-type virus.

A possible application of this tinkering is actually to cure diseases. As bacteria evolve resistance to our strongest antibiotics, “phage therapy” has emerged as a last-line defense. However, bacteria eventually learn to resist natural phages, too.

The researchers explain in a press release:

×

Thank you! One more thing…

Please check your inbox and confirm your subscription.

“Bacterial resistance to antibiotics represents one of the most pressing challenges in modern medicine, with resistant infections killing hundreds of thousands or more annually. Bacteria can quickly evolve resistance to traditional antibiotics, limiting therapeutic effectiveness.

We wanted to see if we could one day design phage therapies that could be resilient against bacterial evolution.”

In the lab, the researchers evolved E. coli that was completely resistant to the standard ΦX174 virus and had it duke it out with the viruses.

When they tried to use the original virus to kill these “superbugs,” it failed. But when they deployed a “cocktail” of their AI-designed phages, something remarkable happened. Within just a few generations, the AI phages recombined and mutated to “pick” the new bacterial locks, successfully wiping out the resistant strains.

The Safety Problem

The researchers took a “safety-first” approach. They deliberately excluded all viruses that infect humans or animals from the AI’s training data. The AI only knows how to build “bacteriophages” — viruses that are harmless to humans but lethal to bacteria.

But this doesn’t mean malicious actors couldn’t use the same method for nefarious purposes.

While ΦX174 is a small virus, the researchers believe this is just the beginning. As AI models get smarter and DNA synthesis gets cheaper, we could eventually design larger, more complex living systems to clean up pollution, produce medicine, or revolutionize how we treat the world’s deadliest diseases. They could also be used to create dangerous pathogens.

We have officially moved from “editing” life to “generative design.” While the researchers were careful to exclude human-infecting viruses from the training data for safety, the blueprint is now clear. Today, it’s a tiny virus hunting E. coli; tomorrow, who knows?

The study was published BiorXiv.