Artificial intelligence · Article
AI-Designed Virus Used to Treat a Drug-Resistant Infection: What We Know
Published:
In brief
Startup Radical Numerics says its Omnii-1 AI model designed a bacteriophage that was given to one patient with a drug-resistant E. coli infection under an FDA compassionate-use pathway. The October 20, 2026 announcement is described as the first AI-designed virus administered to a human, but the findings are not peer-reviewed and outside scientists say they are withholding judgment.
On October 20, 2026, the biotechnology startup Radical Numerics said it had used an artificial intelligence model called Omnii-1 to design a bacteriophage — a virus that infects and kills bacteria — and that the engineered phage had been given to a patient with an antibiotic-resistant Escherichia coli infection. The company describes it as the first time an AI-designed phage has been administered to a human patient.
Why this story is in the news now
Radical Numerics emerged from a period of secrecy on the same day with a $50 million seed round and the public launch of Omnii-1, which it calls a multimodal genome language model. According to the company, Omnii-1 was trained on 28 types of biological data, including DNA, RNA, proteins and epigenetic signals, and can be used to predict disease-causing genetic variants, help design personalized cancer vaccines, and detect emerging biological threats, in addition to designing phages.
The phage treatment was given under a physician-led, single-patient expanded-access route — commonly known as compassionate use — which the U.S. Food and Drug Administration allows for patients who have no other treatment options. Radical Numerics said the FDA authorized the use, but the agency did not comment publicly when the news broke, and the company has not disclosed the patient's identity, the treating physician, or the clinical outcome, citing privacy.
How Omnii-1 designed the phage
According to a technical write-up published by Radical Numerics, researchers started from T7, a well-studied bacteriophage with roughly a 40-kilobase genome that naturally infects E. coli. They used Omnii-1 to generate new variants of gp17, the protein T7 uses to bind to bacterial surfaces, while keeping the rest of the phage genome intact. The goal was a phage that could still bind to and kill the specific antibiotic-resistant bacterial strain isolated from the patient.
The company reports that in laboratory testing, AI-designed variants with 11 to 15 amino-acid changes functioned successfully far more often than randomly generated variants with similar changes. The resulting phage cocktail was tested against bacteria taken from the patient and eliminated the resistant strain in vitro before being administered.
| Variant type | Amino-acid changes | Functional recovery rate |
|---|---|---|
| Omnii-1-designed | 11–15 | 69.9% |
| Random control | 11–15 | 4.8% |
From Evo to Omnii-1: a fast-moving research line
- 2025: Arc Institute researchers used an AI model called Evo 2 to generate about 700,000 candidate phage genomes; 285 were synthesized and tested, and 16 proved viable in laboratory experiments, according to GV Wire.
- October 20, 2026: Radical Numerics launched Omnii-1 publicly alongside its $50 million seed round and announced the compassionate-use treatment of a patient with an AI-designed phage.
Radical Numerics chief executive Eric Nguyen framed the milestone as a step toward individualized medicine, saying that the field is moving away from searching for existing drugs and toward designing treatments matched to a patient's specific biology, according to Yahoo Finance.
What remains unproven
Radical Numerics' technical report on the phage design has not undergone independent peer review, and the company has declined to share the patient's clinical outcome. The FDA did not provide comment to reporters covering the announcement. Phage therapy itself is not new — according to GV Wire it dates back more than a century and is seeing renewed interest as antibiotic resistance grows, with specialized clinics now operating in the United States and elsewhere — but using an AI model to custom-design a phage for a single patient's resistant strain is new.
If it works, that's incredible. My impulse is to be skeptical.
Baym said he would withhold judgment until the treatment runs its course and he can examine how novel the AI-generated viruses actually are compared with existing phages.
A dual-use technology
Radical Numerics says Omnii is also being developed for biodefense, including work with Lawrence Livermore National Laboratory to help detect emerging biological threats, and the company counts former defense officials and Microsoft's chief scientific officer among its advisors. The same genome-design capability that lets AI create a therapeutic phage for one patient's infection is, by the company's own account, a technology it is simultaneously trying to build safeguards against, since AI tools that can design functional viruses could in principle also be misused.
Questions
- What is an AI-designed bacteriophage?
- It is a virus that naturally kills bacteria, modified with the help of an AI model — in this case Radical Numerics' Omnii-1 — so that it targets a specific, often drug-resistant, bacterial strain more effectively.
- Did the FDA approve this as a drug?
- No. The FDA authorized a single-patient compassionate-use (expanded access) treatment for one patient with no other options, which is different from approving the therapy as a general drug.
- Is the AI-designed phage treatment proven safe and effective?
- Not yet in a peer-reviewed sense. Radical Numerics' technical report has not been independently reviewed, the clinical outcome for the patient has not been disclosed, and outside scientists such as Harvard's Michael Baym say they are withholding judgment.
- Which company and AI model are behind this?
- Radical Numerics, a startup founded by members of the team behind the Evo and Evo 2 genome AI models, used its new model Omnii-1 to design the phage.
- Has an AI-designed virus been given to a human before?
- According to Radical Numerics, this is the first reported case of an AI-designed phage being administered to a human patient. An earlier AI model, Evo 2, generated hundreds of thousands of candidate phage genomes in 2025, with 16 later shown to be viable in laboratory tests, according to GV Wire — but none were given to a patient.
Sources
This piece was prepared with artificial intelligence and passed an automated source check before publication. If you see an error, ask for a correction. Request a correction or reply · How we make this
Related
Artificial intelligenceNews
DeepSeek open-sources DeepInfer, an inference tool to cut AI running costs
DeepSeek has released DeepInfer, the inference framework that runs its own production AI services, as free open-source software under the Apache 2.0 license. The tool is built to make serving the company's DeepSeek-V4.1-Flash model faster and cheaper, and the company says sharing it will help other developers lower compute costs and energy use.
Artificial intelligenceArticle
Quantum Computing Explained: How Qubits Work and Why It Matters Now
Quantum computers use qubits that can hold combinations of 0 and 1 at once, letting them explore many possibilities in parallel for specific problems. Google and IBM both reported 2025–2026 milestones that move the technology from laboratory demonstrations toward practical use, while NIST has already issued encryption standards to prepare for the risk quantum machines could pose to today's cryptography.
Artificial intelligenceArticle
How algorithmic bias in AI hiring tools can replicate workplace discrimination
Algorithmic bias in hiring AI happens when models trained on historical workforce data, resumes, or online images absorb and repeat existing patterns of gender, race or age skew. Documented cases — from Amazon's scrapped recruiting tool to a University of Washington study on large language models — show the bias is not hypothetical, which is why New York City and California have introduced specific legal rules for employers using these systems.
Artificial intelligenceNews
OpenAI Upgrades ChatGPT's GPT-6 Astra Pro With Parallel AI Agents
OpenAI announced on October 16, 2026 that GPT-6 Astra Pro inside ChatGPT now lets multiple AI agents work on separate parts of the same task at the same time, instead of one agent completing steps in sequence. The feature is currently limited to ChatGPT Pro subscribers, who can track each agent's progress and open finished work directly in the chat.