Stanford’s Virtual Biotech Deploys 37,000 AI Agents in Drug Discovery

Stanford researchers introduced the Virtual Biotech, a multi-agent artificial intelligence system designed to operate like a biotechnology company, with a structure that includes a Chief Scientific Officer agent and specialized divisions.

The Virtual Biotech system autonomously processed and annotated outcomes from more than 55,000 clinical trials using over 37,000 clinical-trialist agents working in parallel. These agents connected clinical trial results to multi-omic data, including features identified through single-cell RNA-sequencing.

During its analysis, the system identified a molecular signal that was predictive of clinical-trial success. The agents also evaluated drug targets, such as B7-H3 for lung cancer, and proposed specific therapeutic strategies, including an antibody-drug conjugate approach.

In a demonstration, the Virtual Biotech designed a lung cancer therapy that was later independently validated by a pharmaceutical company. This process included linking clinical outcomes to cell-type-specific gene targets and providing recommendations for drug development strategies.

What the numbers show

• Over 37,000 AI agents participated in the system’s analysis.
• The system reviewed more than 55,000 clinical trials.
• Drugs targeting cell-type-specific genes were about 40% more likely to advance from Phase I to II, 48% more likely to reach market, and had 32% fewer adverse events.

The findings from the Virtual Biotech’s work were published in a peer-reviewed Science article on September 17, 2026. The publication detailed the system’s methodology and the outcomes of its large-scale clinical trial analysis.

One of the system’s drug designs was independently confirmed by Merck, according to media reports. The therapy designed by the Virtual Biotech subsequently received FDA breakthrough therapy designation.

The Virtual Biotech’s approach involved organizing thousands of AI agents to mirror the structure and workflow of a real-world biotechnology company. This included specialized agents for different research tasks and a central coordinating agent overseeing scientific operations.

By linking clinical trial outcomes to detailed molecular and cellular data, the system provided insights into which drug targets were more likely to succeed and experience fewer adverse events. This approach enabled the identification of promising therapeutic strategies and supported independent validation by external pharmaceutical organizations.

* This article is based on publicly available information at the time of writing.