AI-designed phage reportedly given to a patient for the first time

San Francisco-based AI lab Radical Numerics announced in October 2026 that bacteriophages designed with its genome model, Omnii-1, have been given to a patient with an antibiotic-resistant bacterial infection. According to the company, the treatment was administered under a physician-led compassionate-use pathway authorized by the FDA. If the claim holds, it would be the first time an organism designed by artificial intelligence has been given to a human as a drug.

The picture remains incomplete, however. The company has shared nothing about the patient's clinical course, the outcome of treatment, or the hospital and physicians involved. The technical report was published on the company's own website and has not been peer reviewed; it is the only source for the case.

Why bacteriophages are back in focus

Bacteriophages are viruses that infect only bacteria and leave human cells alone. Their therapeutic use predates penicillin, and interest has revived as antibiotics lose their effectiveness. Unlike broad-spectrum antibiotics, a phage can target a single bacterial species without harming the rest of the microbiome. The World Health Organization generally considers phage therapy safe.

That precision is also the problem. Each phage recognizes only a narrow set of bacterial surface receptors, so one has to be found that matches the patient's particular infection. Clinicians first search existing phage libraries, then environmental samples, and turn to engineering as a last resort. For a patient with sepsis, this search is a race against time. The scale of the underlying problem is not in dispute: the Global Burden of Disease project estimates that antimicrobial resistance was associated with about 4.7 million deaths in 2021 and could be linked to roughly 40 million deaths between 2025 and 2050.

How Omnii-1 works

Radical Numerics was founded by people from the team behind the Arc Institute's Evo and Evo 2 genome models. The Evo work produced the first phage genomes written entirely by AI: of 700,000 candidate genomes, 285 were synthesized and 16 yielded viable viruses. Despite the heavy attrition, it was a promising proof of concept. Omnii-1 is its successor and, according to the company's technical paper, takes a narrower, more surgical approach: rather than inventing genomes from scratch, the model edits the targeting component.

The team worked on T7, a well-characterized E. coli phage of about 40 kilobases, and rewrote gp17, the gene encoding the tail fiber protein that lets the virus attach to the bacterium. By the company's figures, 69.9 percent of designed variants that differed from the wild type by 11 to 15 amino acids were recovered as viable phage, versus 4.8 percent in a random control group. Of 6,694 variants with measurable activity, 2,161 outperformed the original, and some of those carried 28 amino acid changes.

The patient case and open questions

The company tested its designed phage cocktail against bacteria isolated from a real patient's infection, confirmed in the lab that it killed the bacteria, and then administered it to the patient. The E. coli infection was reportedly resistant not only to antibiotics but to all available phages. The company declined to share the clinical course, outcome, hospital or physicians, citing patient privacy and the risk that even limited clinical detail could identify the patient. Nine days after the announcement, as of October 29, reports aimed at infectious disease clinicians still noted that no clinical course, outcome or peer-reviewed paper had been made public.

The company's own technical paper states plainly that single-patient observations cannot establish safety or efficacy. CEO Eric Nguyen said the team plans to publish the work in a peer-reviewed journal, though the company also said the treatment is proprietary and that it hopes to commercialize it. Independent experts are interested but cautious. Harvard microbiologist Michael Baym, who was not involved, said the result would be incredible if it works, but that he would remain skeptical until treatment is complete and the novelty of the viruses can actually be examined.

The company's commercial logic is clear: phage hunting is slow, and Radical Numerics believes it can cut design time from months to days. Marisa Azad, an infectious disease physician at the Ottawa Hospital who works with the company, describes the conventional search as extremely time-consuming. If the process can be shortened reliably, more patients could be reached in time and treatment offered to more people at lower cost. On the other hand, a model that can design viruses on demand also carries dual-use risk. The company says it is committed to biosecurity, containment and data-sharing safeguards proportionate to the model's capabilities. Engineered phages have been given to patients before; the best-known example is the 2019 treatment of a disseminated, drug-resistant Mycobacterium abscessus infection with modified phages.