Biomedicine · global
First Use of AI-Designed Phages in a Patient? Company Reports Administration, Outcomes Remain Undisclosed
Radical Numerics says a cocktail containing AI-designed phages has been administered to a patient with a drug-resistant infection through compassionate use. Laboratory results showing bacterial killing provide a starting point, but public clinical data on whether the patient benefited are still lacking.
When antibiotics stop working, finding bacteriophages that can target the bacteria in a patient’s body can also be a difficult matching process. Radical Numerics announced on October 20 that phages designed by its artificial intelligence model, Omnii-1, had been administered as part of a cocktail to a patient with a drug-resistant bacterial infection. The company calls this the world’s first use of AI-designed phages in a human, but it has not publicly disclosed the patient’s treatment outcomes.
The treatment was administered through an authorized, physician-led compassionate use pathway, allowing a patient without suitable treatment options to receive an investigational therapy outside a clinical trial. The company said that, in in vitro testing before administration, the preparation eliminated bacterial isolates obtained from the patient. The Register also specifically distinguished these two points: bacterial killing under culture conditions cannot be taken directly as evidence that the infection in the patient’s body has been controlled.
Bacteriophages are viruses that infect bacteria. Those with lytic capabilities can replicate inside bacteria and cause them to rupture. The appeal of this therapy lies in its precision, and so does its difficulty: a phage that can infect one bacterial strain may not be effective against another. Recognizing the bacterial surface is only the first hurdle; the virus must also overcome bacterial defenses and retain its ability to replicate.
Omnii-1 seeks to help researchers narrow this search. According to the company’s technical description, the model combines genomic information with experimentally measured phage–host interaction data to propose sequence variants and prioritize them for testing. AI design can also involve modifying specific regions of existing phages, such as the tail fiber proteins responsible for recognizing bacteria; candidate designs still require synthesis and experimental validation.
The sources of information about the bacterial species infecting the patient need to be distinguished. On October 28, Contagion Live cited reporting by The New York Times stating that the patient was infected with Escherichia coli and that neither antibiotics nor existing phage therapies had worked; the company’s announcement itself did not disclose the species. The Register cautioned that the technical report separately included model testing involving Pseudomonas aeruginosa, which cannot be used to infer that this was the patient’s pathogen.
Contagion Live noted that the company report had not yet undergone peer review, that the company planned to submit its research findings, and that disclosure of clinical details depended on appropriate consent. Citing privacy protection, the company has so far withheld the patient’s clinical course and outcome, making it impossible to determine from public information whether the infection improved or whether the treatment caused adverse reactions.
This case brings AI-designed phages from the laboratory to administration in a human, while leaving the most critical clinical questions unanswered. A single case is insufficient to establish safety or efficacy, and compassionate use authorization does not amount to approval for widespread use. Making such personalized therapies a reliable option will require clinical data that can be scrutinized, along with more comprehensive validation of safety and efficacy.