AI-Designed Viruses Move From Lab Milestone to a Claimed Patient Treatment

Artificial intelligence has taken another step into biology. Scientists have used generative AI to design viruses that do not exist in nature, and a startup now says it has tailored one of these viruses to treat a single patient. The work has raised hopes for new medicines against drug-resistant infections, and it has also stirred concern about misuse.

The viruses in question attack only bacteria, not humans. Researchers describe the achievement as the first set of genomes, or complete genetic instruction books, designed entirely with generative AI. A paper in the journal Science described the breakthrough on August 6, after a non-peer-reviewed version appeared on the preprint server bioRxiv last year.

How Scientists Taught AI to Write Genomes

The team relied on two generative models, Evo 1 and Evo 2, which trained on trillions of nucleotides. Nucleotides are the building blocks of DNA and RNA, and the models drew them from genetic sequences of all sorts of organisms. The systems work somewhat like large language models such as ChatGPT, but they predict and generate patterns of genetic sequences instead of sentences.

NPR science correspondent Katia Riddle explained the analogy on the network’s Short Wave podcast. DNA is written in an alphabet of just four letters, A, C, G, and T, and a model can learn patterns in that language the way ChatGPT learns patterns in text. In her words, you can read a DNA sequence essentially like a book.

To test whether Evo could design a complete genome, the team turned to bacteriophages, viruses that primarily attack bacteria. The researchers gave the models extra training data from Phi X-174, a well-studied bacteriophage that infects only Escherichia coli. After that training, they asked the models to propose new viral genome sequences and introduced those sequences into bacteria roughly 300 times.

A Fast Result and an Early-Morning Surprise

One of the researchers, Brian Hie, works as a scientist at Stanford and previously worked at Facebook. He and colleague Samuel King received the first batch of experimental results in the middle of the night. Hie later described the moment to Riddle.

Hie recalled the first results: “I think Samuel had pulled an all nighter or something like that and got the first batch of around 3:00 or 4:00 AM. I, like, woke up at 6:00 AM a couple hours later and saw the data that was posted onto our group Slack.”

The project moved quickly. Riddle noted that the team finished the work in only one year, far faster than pre-AI science timelines. She also stressed that the researchers took a number of precautions to make the work safe, and that the resulting viruses infect bacteria rather than humans.

Outside experts see the study as a turning point. Patrick Cai, a synthetic biologist at the University of Manchester in England who did not take part in the study, spoke to Carl Zimmer of The New York Times.

“This is an important milestone,” says Patrick Cai, a synthetic biologist at the University of Manchester in England who was not involved in the study, to Carl Zimmer at the New York Times.

Why Bacteria-Killing Viruses Matter

Bacteriophages could help treat bacterial infections, especially because bacteria can develop resistance to existing treatments. Antimicrobial resistance ranks as a major global health threat. Bacteria, viruses, fungi, and parasites sometimes stop responding to medicines, which makes infections hard or even impossible to treat.

The misuse and overuse of antimicrobial treatments, including antibiotics, drive the problem. According to the World Health Organization, bacterial resistance associated with an estimated more than 4.7 million deaths worldwide in 2021. Those numbers explain why researchers look to phage therapy as a possible alternative.

Certain viruses attack bacteria and can sometimes clear antibiotic-resistant infections. AI design could let researchers tailor a virus to a particular infection instead of searching nature for a match. That prospect drives much of the excitement around the new technique.

A Startup Reports Its First Patient

The commercial race has already started. On October 20, 2026, researchers at the startup Radical Numerics announced that they had tailored a virus to attack one patient’s illness and administered it to that patient. The company created the virus in part with its own AI model, Omnii. A visualization the company released shows the tailored virus.

The announcement is a claim from the company, and the sources provide no independent confirmation of the outcome. It involves a single patient, so it cannot show whether the approach works broadly. Even so, it marks a move from computer-designed genomes in the lab toward use in medicine.

Hope and Worry Side by Side

The same capability that promises new medicines also troubles some experts. The research has raised worries about possible dangers of misuse, and experts fear that similar tools could someday help produce harmful pathogens. The study’s authors limited their risk by working with viruses that attack only bacteria and by taking safety precautions.

Artificial intelligence already helps discover drugs, modify key machinery in microbes, and decode the human genetic instruction book. Designing whole viruses from scratch extends that reach to a new level. The coming months will show whether the tailored-virus approach can deliver real treatments, and how researchers and regulators balance its benefits against its risks.