A human subject was injected with an AI-designed virus aimed at killing antibiotic-resistant bacteria, so let’s see how it goes

When you purchase through links on our articles, Future and its syndication partners may earn a commission.

• San Francisco lab Radical Numerics says its Omnii-1 model engineered bacteriophages that were delivered to a patient with a drug-resistant infection, a world first

• The phages were built by rewriting the T7 tail fiber gene, the protein the virus uses to latch on to bacteria.

• No one outside the company knows if it worked, because the results have not yet been published, peer-reviewed, or disclosed.

Radical Numerics announced that its AI-designed viruses (bacteriophages) were administered to a sick person, but did not provide much more information about what happened next.

The San Francisco AI lab announced in October 2026 that bacteriophages engineered by its new genomic model had been administered to a human patient with an antibiotic-resistant bacterial infection, under a physician-led compassionate use pathway authorized by the FDA.

If this claim is true, it is the first time that any organism designed by AI has been administered to a person as a medicine.

This is not a new phenomenon, but one that could benefit from help from AI

Bacteriophages are viruses that infect bacteria and completely ignore human cells. They have been used as therapy for more than a century, before penicillin, and saw a resurgence of interest when antibiotics began to fail.

The World Health Organization considers phage therapy to be generally safe, and unlike broad-spectrum antibiotics, a phage can target one bacterial strain while leaving the rest of your microbiome alone.

This precision is also the problem: Each phage recognizes a narrow set of bacterial surface receptors, so treating a patient means finding one that matches their specific infection. Clinicians look at existing phage libraries, then environmental samples, then give up and attempt engineering as a last resort. For a person with sepsis, this research is not an academic exercise but an important life and death situation.

The magnitude of the underlying problem is not in dispute. The Global Burden of Disease Project estimated that antimicrobial resistance was associated with approximately 4.7 million deaths in 2021, with projections indicating that between 2025 and 2050, almost 40 million deaths could be attributable to it.

Radical Numerics was founded by members of the team behind Evo and Evo 2, the Arc Institute’s genomic models that generated the first phage genomes written entirely by AI. This previous work produced 700,000 candidate genomes; 285 were synthesized and 16 produced viable viruses. It was a proof of concept with a success rate that was both punishing and promising.

Omnii-1 is the follow-up, and the company’s technical description describes a narrower, more surgical approach. Rather than inventing entire genomes, the model modifies the targeting component. The team worked with T7, a well-characterized 40-kilobase E. coli phage, and rewrote gp17, the tail fiber binding to the receptor.

The numbers the company is willing to share are truly interesting. With 11 to 15 amino acid changes compared to wild type, 69.9% of engineered variants were recovered as viable phages, compared to 4.8% for random controls. Of 6,694 variants with measurable activity, 2,161 beat the original, including some carrying 28 amino acid substitutions.

A promising piece, but untested and unpublished

The company tested its designed cocktail against bacteria isolated from an actual patient’s infection, confirmed that it cleared that isolate in vitro, and then administered it. The patient reportedly had an E. coli infection that had resisted not only antibiotics but also all available phages, which is indeed the worst situation one can find themselves in when it comes to infectious diseases.

Radical Numerics has since declined to describe the patient’s clinical course, findings, hospital or doctors, citing patient confidentiality and the fact that even limited clinical details could identify the patient.

The company self-published the technical report on its website and has not had it peer-reviewed, but has separately indicated that the treatment is proprietary and that it hopes to commercialize it.

This silence continued until October 29, nine days after the announcement, while commercial coverage of infectious disease clinicians still reported no disclosed clinical developments, no results, and no peer-reviewed publications. The technical writing remains the only account of the affair.

In fairness to Radical Numerics, its own article clearly states that observations on a single patient cannot establish safety or effectiveness, which is more conservative than most product launches do. CEO Eric Nguyen also said the company intends to release its publications through peer review.

Outside experts have landed somewhere between intrigued and unconvinced, which is the right place to land. Harvard microbiologist Michael Baym, who was not involved in the study, said the outcome would be incredible if it worked, but added that his instinct was to remain skeptical until the treatment was complete and the novelty of the viruses could actually be examined.

The business logic is much easier to establish; Phage research is slow and Radical Numerics believes it can reduce conception to days rather than months.

Marisa Azad, an infectious disease clinician at the Ottawa Hospital who collaborates with the company, described conventional research as time-consuming, and she’s not wrong. If AI can reliably shorten this loop, it will change who gets treated in time, allowing access to a more cost-effective and wider audience for said treatments.

There is also cause for concern: a model capable of designing viruses on command is, structurally, a model capable of designing viruses on command. Radical Numerics directly recognizes this by committing to biosecurity, containment and data sharing measures appropriate to the capabilities of the model.

This is the same dual-use problem that affects all successful biological design tools, and that AI labs are publicly wrestling with as their systems improve in biology.

Engineered phages have reached patients before, with the most famous case being a 2019 case in which engineered phages treated a disseminated drug-resistant Mycobacterium abscessus infection. What is new here is not the therapeutic concept but the design process, and the governance issues related to the AI-driven process, not the concept itself.

AI has gone from reading biology to writing it and giving the result to a doctor. This is a different category of claims than an AI model proposing a cancer hypothesis or outperforming specialists in diagnosing rare diseases, both of which keep the human firmly in the loop before anything enters the bloodstream.

But for now, a very clever experiment was conducted on a very sick person, and we still don’t know how it happened or when anyone decides to tell us. However, if this comes to fruition, it will have far-reaching consequences for an industry that is only expected to grow over the coming decades.