First AI-Designed Virus Therapy Administered to a Human Patient
AI is no longer just writing code. It is writing biology, and the first AI-designed virus has officially been given to a human patient.
Research · Source: BusinessWire (company announcement)
What happened
Radical Numerics built an artificial intelligence model called Omnii. It does not write software or generate images. It writes DNA, RNA, and proteins to build general biological intelligence. The startup used this model to design a custom virus, known as a phage, to attack a deadly bacterial infection. Instead of spending months searching for the right naturally occurring phage, the AI personalized a virus to match the exact bacterial infection of a single individual.
The FDA greenlit the experimental treatment under its single-patient expanded access pathway. This is commonly known as compassionate use. Doctors administered this AI-generated phage therapy to a human patient who had no comparable alternative. Before the injection, lab tests showed the AI-designed cocktail successfully killed the bacteria taken directly from the patient. Radical Numerics claims this marks a massive milestone for artificial intelligence in medicine.
We do not know if the patient survived or recovered. Radical Numerics refused to share clinical outcomes or specific dates, citing patient privacy and clinical trust. Chief Executive Officer Eric Nguyen stated they intend to publish findings later. The underlying report posted on the company website has not been peer-reviewed. The FDA also declined to provide a comment on the record regarding the specific trial.
Key facts
- 2025 — Year AI startup Radical Numerics was founded
- Omnii — Name of the genome language AI model used
- One — Number of patients treated under FDA compassionate use
Why it matters
Biology is officially becoming a search and generation problem. We are moving from discovering drugs in nature to generating them on demand using artificial intelligence. If language models can learn the complex grammar of molecular landscapes, biotech becomes a pure software engineering discipline. Founders no longer need to rely purely on trial and error in wet labs. They can compute the exact biological mechanism needed to kill a specific drug-resistant bacteria. This shifts the bottleneck from discovery to compute and clinical testing.
The regulatory ceiling is also shifting rapidly. The FDA allowing an AI-generated virus into a human body sets a massive precedent for the entire healthcare industry. Startups building biological AI now have a proven regulatory pathway to test personalized, single-patient therapies. This compassionate use authorization proves that regulators are willing to accept AI-designed biological agents when patients face life-threatening conditions. The barrier to entry for testing AI therapeutics just got significantly lower.
For builders
Personalized medicine as a service
Radical Numerics proved you can generate a custom therapy for one specific patient. Startups can now build infrastructure to support single-patient FDA applications and rapid clinical deployment. Pharmaceutical companies will pay a massive premium for platforms that automate this highly personalized pipeline.
Biodefense and risk modeling
The same artificial intelligence that cures infections can evaluate pandemic risks and build biodefense mechanisms. Governments and defense contractors desperately need software to simulate and defend against biological threats. Builders who create evaluation tools for biological AI will find lucrative defense contracts waiting for them.
Biological data pipelines
Models like Omnii require massive amounts of clean data across cellular systems and proteins. Engineering teams that can scrape, clean, and structure biological data will become highly valuable. AI labs will pay heavily for exclusive access to high-quality microbial ecosystem datasets.
My take
AI crossing the barrier from digital text to human biology is the most important shift of our decade. We are no longer just writing code, we are writing life. But running aggressive PR campaigns before publishing peer-reviewed clinical outcomes is a dangerous game. If this AI-designed virus fails, the regulatory backlash will set biological AI startups back by years.
Original reporting: BusinessWire (company announcement). This is my rewrite and opinion.