Bacteriophages, viruses that infect and kill bacteria, were administered to patients before penicillin existed. A hundred years later, with drug-resistant infections killing more than a million people worldwide each year, not one phage therapy has been approved by the U.S. Food and Drug Administration.
That gap is not due to a lack of trying. Trials are running on several continents, and last year produced the field's clearest randomized signal yet: a phage cocktail that measurably shrank diabetic foot ulcers where bone infection had set in. The obstacles are stranger and more mundane than the biology, which is part of why artificial intelligence has become such an object of interest here.
One Randomized Signal, in a Wound That Rarely Heals
Diabetic foot osteomyelitis is a brutal problem. The ulcer sits over infected bone, standard care frequently fails, and amputation is a real endpoint.
BiomX tested a phage product called BX211 against placebo in a randomized, double-blind, multicenter Phase 2 trial involving 41 patients with Staphylococcus aureus infections confirmed by bone biopsy. Twenty-six received intravenous plus topical phage in week one, then weekly topical dosing through week 12. Fifteen received placebo. Everyone got standard care, including antibiotics.
The results, first reported in March 2025 and later presented at IDWeek, showed that the treated group's ulcers separated from the placebo group starting at week 7, with a difference greater than 40% by week 10. Percent area reduction reached a p-value of 0.046 at week 12. Ulcer depth improved in patients whose wounds reached bone at baseline, and the treatment limited ulcer expansion.
Two caveats go together. The primary endpoint was measured at week 13, with a p-value of 0.052, which is not statistically significant. And the abstract states directly that the statistical significance was not adjusted for multiple comparisons. In a 41-person trial testing several endpoints, that matters.
The Bottleneck Is Not the Virus, It Is the Match
Antibiotics are blunt instruments, which is both their weakness and their operational advantage. A phage is the opposite. It is exquisitely specific, often to particular strains within a species, which spares the rest of a patient's microbiome but creates a logistics problem.
BX211 is not one drug. It is phages drawn from a pre-established phage bank and individually matched to each patient's own S. aureus isolate. For a critically ill patient, that matching has to happen fast, and often it does not happen at all because nothing in the collection fits.
This is where AI enters the picture in a way that has drawn less attention than genome design. A narrative review published in Antibiotics in June cataloged the computational work now aimed at the pipeline itself: mining metagenomic data for new phages, annotating their genomes, predicting which bacterial hosts a given phage can infect, selecting a phage for a specific patient, optimizing cocktails, and designing phage plus antibiotic combinations. The authors frame these as approaches that may help, not solutions shown to work in patients.
Evidence that phages help is genuinely mixed. A randomized trial in men undergoing prostate surgery found bladder-instilled phages non-inferior to standard antibiotics but not superior to placebo. A 2026 systematic review in the International Journal of Infectious Diseases covering 51 cystic fibrosis patients with drug-resistant infections found microbiological improvement in 69% and lung function improvement in 74% of those measured, drawn from 19 studies, of which only two were randomized trials.
A Nebulizer, a Safety Review, and a Canceled Trial
The field's setbacks illustrate how unglamorous the barriers can be.
BiomX's other lead program, BX004, was an inhaled phage cocktail for chronic Pseudomonas aeruginosa infection in cystic fibrosis, tested in a Phase 2b trial of about 60 patients. In August 2025, the FDA placed a clinical hold on the U.S. portion. The hold was not about the phages. It concerned the third-party nebulizer used to deliver them, and the agency raised no concern about the drug itself.
Then the drug did become the problem. An independent data monitoring committee reviewed adverse events and initially recommended continuing with revised dosing. Weeks later, after further internal analysis identified unexpectedly high rates of adverse events, the company discontinued the trial in December 2025 and shifted its focus to a fixed phage cocktail for diabetic foot infections, pending funding.
Regulators elsewhere have started building scaffolding. The European Pharmacopeia adopted a general chapter on medicinal products for phage therapy that took effect in January 2025, and the World Health Organization's 2024 list of priority bacterial pathogens has strengthened the case for alternatives. A 2026 review of the regulatory landscape describes a field still assembling basic quality and governance standards, with access routes rather than approval routes doing most of the work.
How Patients Actually Get Phages Today
For an American patient with an infection that antibiotics cannot clear, the practical options are limited. A small number of academic centers provide phage therapy on a case-by-case basis under FDA expanded access, using phage banks and requiring a matched isolate. It is arranged through an infectious disease physician, not requested off a website.
Phage products marketed directly to consumers are not evaluated for safety or effectiveness, and anyone with a resistant infection should go through infectious disease consultation and culture-guided treatment rather than pursuing phages independently.
The realistic reading is that phage therapy has moved from anecdote toward evidence, but has not yet arrived. One modestly sized trial with unadjusted statistics and a missed primary endpoint is a beginning, not a verdict.
Key Questions Answered
What is phage therapy?
Treatment using bacteriophages, viruses that infect and kill bacteria. They cannot infect human, animal or plant cells.
Is any phage therapy FDA-approved?
No. Trials are underway in several countries, but no phage product has received approval in the United States, and no randomized trial has yet demonstrated the efficacy required for approval.
What did the strongest trial show?
In 41 patients with diabetic foot osteomyelitis, a matched phage cocktail plus standard care shrank ulcers significantly more than placebo at week 12. The result did not reach statistical significance at the week 13 primary endpoint and was not adjusted for multiple comparisons.
Why is matching phages to patients so hard?
Phages are strain-specific. A phage that kills one patient's Staphylococcus aureus may do nothing against another's, so treatment requires searching a phage bank for a match against a fresh isolate.
What is AI actually being used for here?
Beyond the recent headlines about designing phage genomes, computational work is focused on finding new phages, predicting which bacteria they can infect, and selecting or combining them for individual patients. None of this has been validated in patients.
How would a patient access phage therapy now?
Through an infectious disease physician at one of the few centers offering it under FDA expanded access. Consumer phage products are not evaluated and should not be used.
