Antibiotics are failing us. When superbugs laugh off conventional drugs, doctors face a terrifying wall. Now, a startup called Radical Numerics just cracked open a radical door by treating a human patient with a virus built entirely by artificial intelligence.
Forget reading DNA. We are officially in the write era.
The Crisis of Drug-Resistant Superbugs
Traditional antibiotics act like sledgehammers. They wipe out infections, but they also torch your gut microbiome and leave room for resilient, drug-resistant bacterial strains to thrive. Year after year, pathogens evolve faster than pharma companies can push new molecules through clinical trials.
Enter bacteriophages. These are viruses that exclusively hunt and destroy bacteria. For decades, phage therapy was treated as a fringe or last-ditch option because finding or engineering a natural phage that matches a specific patient's unique bacterial isolate took months of tedious lab work.
Time is a luxury you don't have when septic shock is setting in.
How Omnii-1 Changes the Equation
Radical Numerics launched its new biological AI system, dubbed Omnii-1, to solve this bottleneck. Built by the team behind pioneering genome language models like Evo and Evo 2, Omnii-1 doesn't just predict how proteins fold. It models complex phage-host interactions and designs functional viral genomes from scratch.
When an infection strikes, the workflow begins with a simple bacterial isolate from the patient. Instead of screening thousands of natural phages hoping for a hit, the AI generates and prioritizes synthetic phage variants optimized to target that exact bacterial strain.
In recent evaluations, Omnii-1 achieved remarkable accuracy, identifying active phages across over ninety percent of eligible strains with just a couple of top recommendations. That speed transforms phage therapy from a months-long research project into a rapid clinical response.
What Happens When AI Writes Biology
Using an AI-generated virus on a human patient sounds straight out of a sci-fi thriller. Critics point out the obvious hazards: if a machine learning model can design a virus to cure a bacterial infection, the same underlying architecture could theoretically be inverted.
Radical Numerics approaches this dual-use risk as an architectural problem rather than an afterthought. But regardless of the safety debates, the clinical reality is here. We are no longer limited to the catalog of organisms that evolution happened to test over millions of years.
You can design custom biological agents tailored to a single person's genetic landscape.
The Road Ahead for Personalized Phage Design
Getting an AI-designed phage into a single human patient is a massive milestone. Scaling that success into a routine clinical tool requires closing the loop between digital generation and wet-lab validation. Every synthesized variant that kills a target pathogen feeds back into the model, sharpening its predictive fitness for the next round.
Traditional medicine relies on standardization. Biology is infinitely variable. By combining genome language models with rapid synthesis, startups are building a future where treatments are compiled on demand.
The next time a superbug beats our best drugs, we won't wait for evolution. We will write the countermeasure.
