In a development that could reshape how doctors fight antibiotic-resistant infections, the biotech company Radical Numerics reported in October that it had used an AI-designed virus to treat a patient. The virus, a bacteriophage, was created using a model called Omnii, according to the company. The work was part of a collaboration with another biotechnology firm, Tabula Bioscience, which led the broader effort, Radical Numerics said.

Bacteriophages, or phages, are viruses that naturally attack bacteria rather than human cells. For decades, antibiotics have been the standard treatment for bacterial infections like strep throat and pneumonia. But as bacteria have evolved resistance, the need for alternatives has grown. The U.S. Centers for Disease Control and Prevention estimates that more than 9.3 million drug-resistant infections occur in the country each year. Radical Numerics warns that antibiotic resistance could claim nearly 40 million lives globally between 2025 and 2050.

The challenge with phage therapy has always been finding a phage that can latch onto the specific bacterial strain making a patient sick. Each strain has a unique set of surface proteins, and a phage needs matching molecular machinery—often described as "legs"—to grab hold. Scientists can tweak the genes that code for these legs, but figuring out which changes will work is difficult.

That is where AI comes in. The Omnii model learns what a given organism's DNA typically looks like by comparing many sequences. When given a stretch of DNA with a gap, the model can predict what nucleotides belong there. In designing a phage, scientists keep the parts that already work and ask the AI to suggest changes to the DNA that codes for the anchoring mechanism. According to Northeastern University biotechnology professor Sudhakaran Prabakaran, who spoke with Northeastern Global News, the model might produce several candidate designs, and "maybe two of those options result in a viable organism" that can kill the target bacteria.

Radical Numerics has not disclosed how the AI-designed virus was administered to the patient, citing privacy concerns. In general, phage therapy can involve IV delivery, topical application to a wound, or inhaled treatment, with repeat doses often needed. The treatment has not received formal approval from the U.S. Food and Drug Administration, but the agency authorizes experimental therapies on a case-by-case basis for patients who have exhausted other options. The company's report has not undergone peer review, and the patient's outcome has not been shared.

Prabakaran said he is "cautiously optimistic" about the development but raised concerns about creating synthetic organisms. As a phage replicates inside a patient, mutations can accumulate, potentially undermining safeguards designed to keep the virus from causing unintended harm. Alexander Sulakvelidze, president and CEO of the Maryland-based phage company Intralytix, expressed excitement about the work's potential but flagged risks including unreliable AI predictions and the possibility of using the technology to create harmful biological agents. "AI should complement, rather than replace, scientific expertise and laboratory testing," Sulakvelidze told Northeastern Global News.

If AI-designed phages prove effective, they could offer a way to treat infections that no longer respond to antibiotics. Doctors could also design phages that target harmful bacteria while sparing beneficial microbes, which are often disrupted by broad-spectrum antibiotics. Minmin Yen, CEO of the Boston-based startup PhagePro, said the approach could also be scaled to address global disease burdens like cholera, extending beyond personalized medicine to entire communities.

Still, significant questions remain. If each patient requires a custom-designed virus, it is unclear how the FDA would regulate such treatments. "All of these questions are open," Prabakaran said.