Radical Numerics 的 Omnii-1 設計噬菌體首次施用於患者

Workflow for personalized bacteriophage development using Radical Numerics’ Omnii-1 model. (Source: Radical Numerics)

An artificial intelligence (AI)-designed virus has been administered to a patient for the first time in the United States. The development is drawing attention to whether AI can accelerate the creation of personalized treatments targeting individual patients’ pathogens.

On Oct. 22 local time, San Francisco-based AI company Radical Numerics announced that a bacteriophage cocktail designed using its proprietary AI model, Omnii-1, had been administered to a patient under a compassionate-use treatment protocol. According to the company, this marks the first human administration of AI-designed bacteriophages.

If Radical Numerics’ clinical effort succeeds, AI could enable the rapid development of personalized treatments by designing or selecting bacteriophages effective against individual patients’ bacterial infections. The approach could make a significant contribution to addressing antibiotic-resistant bacteria, commonly known as superbugs, which have emerged as a major public health threat.

Antibiotic resistance occurs when bacteria survive and multiply despite exposure to antibiotics. When existing antibiotics become ineffective, alternative drugs must be found, but treatment options diminish as resistance spreads across multiple medications. Infections caused by multidrug-resistant bacteria can make diseases such as pneumonia and sepsis difficult to treat, while posing serious risks to patient safety and medical procedures where infection prevention is critical, including surgery, organ transplantation and cancer treatment.

Bacteriophages, which attack specific bacteria rather than human cells, have therefore attracted attention as an alternative treatment for antibiotic-resistant infections. Combining multiple phages to broaden the range of bacteria targeted is known as a phage cocktail. While bacteria may develop resistance to a single phage, combining several phages could reduce that risk and expand the treatment’s effectiveness against different bacterial targets.

Under conventional methods, identifying suitable bacteriophages for a patient and preparing them for administration can take weeks to months. AI could shorten the phage design process itself to just a few days. However, the time required before actual administration still depends on efficacy validation, production and safety testing.

Radical Numerics has not disclosed the treatment outcome or the type of infection, citing patient privacy and the limitations of a single case. Nevertheless, the company described the administration as a milestone in bringing AI-designed treatments into clinical use and expressed expectations that it could help establish a process for designing personalized phage therapies.

If Radical Numerics’ clinical effort proves successful, it would mark a transition for biological AI, or Bio-AI, from laboratory simulation and validation to actual human administration under compassionate-use treatment. Stanford University and the Arc Institute also conducted joint research in 2025 on accelerating phage design using AI.

Drug development typically takes around 10 years, with costs estimated to exceed $2 billion when failed development efforts are included. Even so, only about 10% to 12% of drugs entering Phase 1 clinical trials ultimately receive approval. Expectations that AI could reduce development time and costs have therefore fostered close ties between the pharmaceutical and AI industries.

Recently, US pharmaceutical giant Eli Lilly partnered with AI drug discovery company Insilico Medicine, while AstraZeneca announced a multiyear collaboration with AI drug developer insitro to develop treatments for metabolic dysfunction-associated steatohepatitis (MASH) and cardiomyopathy. Insitro uses its Virtual Human platform, which combines laboratory-grown cell models with machine learning, to identify the underlying causes of diseases and discover new drug targets.

German pharmaceutical company Bayer has also partnered with AI drug developer Iambic to discover small-molecule drugs. The collaboration aims to use Iambic’s AI platforms, Enchant and NeuralPLexer, to identify new drug candidates that act on disease targets difficult to address with conventional technologies.

Danish pharmaceutical company Novo Nordisk is working with Anthropic and OpenAI on drug discovery. OpenAI also recently acquired Orchestra Bio, which provides a platform for managing and coordinating drug development processes at biotechnology companies. OpenAI said it plans to combine the platform with its specialized life sciences models and tools to develop AI agents that connect research findings with subsequent experiments, clinical trial preparation and scheduling.