Radical Numerics Claims AI-Designed Phage Used for E Coli Patient

A startup has used an AI-designed virus to treat a human patient with a drug-resistant bacterial infection, moving from searching for natural cures to digitally engineering them. Radical Numerics reports that its Omnii-1 model designed a bacteriophage specifically to target an E. coli infection that had resisted both standard antibiotics and existing natural phage therapies.

This development matters because it addresses the “superbug” crisis—the rise of antibiotic-resistant bacteria that leave doctors with few options for life-threatening infections. While phage therapy (using viruses to kill bacteria) is not new, the process of finding a specific phage that matches a specific bacterial strain is traditionally slow and difficult. By using a genome language model to design the virus, Radical Numerics claims it can bypass this search and create a precision tool in a matter of weeks.

Omnii-1 designs a microscopic lunar lander

The technical process involved modifying the T7 bacteriophage, specifically the DNA encoding a receptor-binding tail fibre protein known as gp17. The researchers altered the protein that allows the virus to attach to its target while preserving the rest of the viral genome.

Radical Numerics Claims AI-Designed Phage Used for E Coli Patient — Photo: Yahoo Finance

The AI tool, Omnii-1, was trained on datasets of experimentally measured phage-host interactions from both public libraries and the company’s own data. To prove its efficacy before human application, the tool was tested against 37 strains of E. coli that it had never encountered during training. Radical Numerics reports that Omnii-1 outperformed two other specialized machine learning systems, reaching 91.4% of eligible strains using just two phages.

Once the AI-designed candidates were identified, they were ranked by predicted fitness and tested in a lab. The resulting phage cocktail successfully eliminated the patient’s bacterial isolate in vitro—meaning it worked in a petri dish—before being administered to the patient under an FDA compassionate use designation.

The gap between lab success and clinical proof

Despite the technical achievement, the medical community is treating the news with heavy skepticism. The findings were released in a technical report on the company’s own website on October 20, 2026, rather than in a peer-reviewed scientific journal. Radical Numerics has not disclosed the patient's clinical course or the final outcome of the treatment, citing the need to protect patient privacy.

Instagram

“if it works, that’s incredible,” Harvard microbiologist Michael Baym told the NYT, but his “impulse is to be skeptical.”

This caution is rooted in the complexity of the human body. Evolutionary microbiologist Brandon Berryhill warned that laboratory experiments cannot reliably predict how a synthetic phage will behave during a complex human infection. There is also the risk of recurrence; doctors typically must monitor phage therapy patients for months to ensure the infection does not return.

A broader shift toward biological AI

The stakes extend beyond individual patients. Omnii-1 is a multimodal genome language model capable of analyzing DNA sequencing data in near real-time. This could potentially move biological threat detection out of the lab and directly into emergency rooms or battlefields using portable DNA sequencers.

Photo: One promising approach to targeting unwanted bacteria is phage therapy, which uses naturally occurring and engineered — Photo: Universities of Wisconsin

However, the “AI-first” approach to medicine brings its own set of frictions. The massive data centers required to power these biological models increase electricity and water demand, creating a tension between the promise of medical breakthroughs and the environmental cost of the computing power required to achieve them.

The risk of startup medicine

The current situation highlights a recurring tension in modern biotech: the collision of investor expectations and regulatory evidence. Because the treatment is proprietary and intended for monetization, the lack of transparent clinical data leaves independent researchers unable to verify if the AI-designed phage actually cured the patient or simply performed well in a lab.

An AI-designed virus is administered to a patient to protect against a multi-resistant bacterium.

If the results are eventually validated through peer review, it could signal a new era of personalized medicine where treatments are written in code and printed in DNA. If they are not, it may serve as another example of startup claims outpacing clinical reality.

Whether this specific patient recovered remains an open question, and until Radical Numerics provides consent-based clinical details, the world is left with a promising technical report and a very quiet patient.