First Patient Injected With an AI-Designed Virus to Fight Superbugs

Targeting deadly infections when all antibiotics fail

The experimental treatment deployed synthetic bacteriophages engineered to hunt a specific, drug-immune strain of bacteria.

Bacteriophages are ancient viruses that hunt bacteria exclusively while ignoring mammalian tissue. For over a century, clinicians have toyed with phage therapy as an alternative to chemical treatments. Yet natural phages are hyperspecific: finding one from nature that latches onto a patient's unique infection can take agonizing months that a patient in septic shock simply does not have.

That timing bottleneck is where generative biology steps in. Radical Numerics, founded by researchers behind the pioneering Evo genome models, used its proprietary Omnii-1 system to modify a classic T7 virus. Instead of attempting to invent a viable synthetic organism from scratch, the team targeted gp17, the tail fiber protein the virus uses to grip onto bacterial walls.

The model introduced up to 28 precise amino acid substitutions to alter the target mechanism. According to technical documentation released by the startup, 69.9% of the synthetic variations generated were viable, functional phages, drastically beating random biological mutation controls.

Radical Numerics Omnii-1 phage intervention details

The bioethics debate over secrecy and dual-use models

The clinical intervention took place under an emergency, physician-led expanded access pathway authorized by the US Food and Drug Administration. The recipient suffered from a resistant E. coli strain that had shrugged off every standard antibiotic and every naturally occurring phage available.

What happens next is where the scientific community grows uneasy. Radical Numerics self-published its technical design data online, but it has not shared the clinical outcome, the treating hospital, or whether the infection cleared. While CEO Eric Nguyen cited patient confidentiality and promised peer-reviewed publication down the road, infectious disease specialists remain split between awe and frustration.

High-stakes deployments of generative algorithms into medical decisions have already proven legally contentious, as seen when OpenAI was sued for wrongful death over ChatGPT medical advice. But synthesizing living biological agents escalates the accountability problem from bad text outputs to active physical matter.

Beyond clinical transparency lies an inescapable biosecurity dilemma: software capable of designing targeted viral proteins to kill bacteria operates on the same mathematical principles required to craft harmful biological agents. While the startup insists it maintains strict synthesis screening, the precedent of putting generative biological agents into human bloodstreams raises the stakes for global oversight.

Availability and patient access

This intervention does not represent an approved, widely available commercial drug. It was cleared exclusively for a single, terminal clinical case inside the United States through the FDA's compassionate-use rules, which allow unapproved experimental agents when no alternatives exist.

Regulatory bodies in the UK and European Union maintain similar emergency pathways for unlicensed therapies, but broad clinical trials for AI-synthesized phages remain several years away.

Generative biology has crossed from computational curiosity to emergency medicine in less than three years. Whether Omnii-1 succeeded in saving its patient or not, biological engineering now has an entirely new playbook—one that regulators and bioethicists will spend the next decade trying to catch up with.