Could AI Kill Superbugs? Company Says AI-Designed Phages Given to First Patient, but Outcome Remains Unknown

Radical Numerics says an AI-designed phage treatment was administered to one patient under a compassionate-use pathway authorised by the US Food and Drug Administration (FDA). Reports say that the patient had an E. coli infection that resisted antibiotics and available natural phages.

The start-up Radical Numerics reported the treatment in a technical report published on its website. The company has not disclosed the patient's clinical outcome. Scientists also caution that it remains unclear whether AI-designed viruses can reliably treat dangerous infections in humans.

The case offers a potential new approach to antibiotic resistance, but the available evidence does not yet establish whether the technology will work beyond laboratory testing.

AI Modified the Virus's Bacterial Targeting Mechanism

Radical Numerics used its proprietary multimodal genome model, Omnii-1, to design the treatment. Researchers modified the DNA encoding a receptor-binding tail fibre protein, known as gp17, in the T7 bacteriophage.

Bacteriophages are viruses that infect bacteria. In this case, the researchers altered the protein that allows the virus to attach to its bacterial target. The company says researchers preserved the broader viral genome. The resulting engineered phage variants were tested against the patient's bacterial isolate.

Laboratory tests found that the AI-designed phage cocktail eliminated the patient's bacterial isolate in vitro. However, success in a laboratory does not establish that the treatment will control an infection inside a human body.

AI Could Reduce Virus Design From Months to Days

Traditional phage therapy involves searching environmental samples for naturally occurring viruses capable of infecting a patient's specific bacterial strain. Finding a suitable match can take weeks or months.

According to the company, AI can accelerate the design and prioritisation of candidate phage variants, potentially reducing the computational design stage to days. Synthesis, laboratory validation, production and safety testing require additional time.

Omnii-1 can also learn interaction patterns across bacterial species, including E. coli and Pseudomonas aeruginosa. This could help researchers develop treatments for pathogens that have received less scientific attention.

Combining multiple phages into a cocktail may also broaden the range of bacteria targeted and reduce the risk of resistance emerging against a single virus. However, AI-assisted design does not eliminate the need for laboratory synthesis, safety testing and purification.

Scientists Question the Evidence Behind the Claims

The findings were published as a technical report on the company's website, not in a peer-reviewed scientific journal. This leaves independent researchers with limited information to assess the treatment.

Radical Numerics cited confidentiality agreements and patient privacy protections as reasons for withholding the clinical outcome and specific treatment details.

Evolutionary microbiologist Brandon Berryhill warned that laboratory experiments cannot reliably predict how a synthetic phage will behave during a complex human infection.

He also noted that producing active phages computationally does not explain why a treatment succeeds or fails. Scientists have yet to establish all the biological rules governing these outcomes.

Microbiologist Michael Baym reportedly reserved judgment pending trial results and an independent assessment of the synthetic phages' novelty. Doctors must also monitor patients for months after phage therapy to check whether infections return.

FDA expanded-access pathways can allow patients with serious or immediately life-threatening conditions to receive investigational medical products outside clinical trials when the applicable eligibility criteria are met. However, results from a single patient cannot establish a treatment's safety or effectiveness for broader clinical use.