# AI Agent Controls Brain Injection in Live Mouse as Exin Tests Autonomous Lab Equipment

Exin Therapeutics connected an AI agent to a robotic surgical system for an experimental gene therapy procedure. The injection was completed under human supervision, but the treatment's effectiveness remains unknown.

- Canonical: https://reviewblip.com/news/exin-therapeutics-ai-agent-mouse-brain-injection-2026/
- Published: 2026-10-31T11:30:58.680Z
- Updated: 2026-10-31T11:30:58.680Z
- Author: ReviewBlip
- Category: Computing
- Type: News

![Exin Therapeutics' experimental laboratory setup with computer display, microscope, robotic positioning equipment, and a mouse undergoing a research procedure.](https://reviewblip.com/editorial-images/cc27a41d1ad35a3725009b6bb91d5123a1ef9e7780d326384f558039e44969cf)

Image credit: Courtesy of Exin Therapeutics, Inc.

An AI agent has controlled a robotic system to deliver an experimental gene therapy into a living mouse's brain, according to new reporting published by WIRED on October 31. The demonstration, conducted by San Francisco-based Exin Therapeutics, shows how AI systems could move beyond analyzing scientific data and begin operating physical laboratory equipment. It was an experimental research procedure, not a clinical treatment or demonstration of fully autonomous surgery.

Exin CEO and cofounder Gabriel Ocana Santero told WIRED that his team connected an AI agent to a custom surgical platform equipped with a camera, robotic positioning hardware, and an injection system. The experiment was designed to investigate a potential treatment for seizures in a mouse with epilepsy. The agent selected a therapy from 19 available options and determined the target brain region, injection coordinates, delivery volume, and speed. A veterinary surgeon performed the initial craniotomy before the agent controlled the injection equipment.

The researchers encountered a significant difference between AI models. According to Santero, Anthropic's Claude Opus 5.5 initially handled planning but declined to proceed with guiding the needle into the animal's brain. The team subsequently used Claude Sonnet 5, which completed the procedure. The account illustrates how differences in model behavior can become operational constraints when software interacts with physical systems rather than simply generating text.

Human oversight remained central throughout the experiment. Researchers required approval of the injection plan, installed an emergency motor-disconnection control, and restricted the agent from modifying its underlying equipment code. When a coordinate-system problem interrupted the procedure, the agent identified the issue but could not correct the software itself. A separate AI agent was used to implement the fix, extending the procedure from a typical duration of approximately 45 minutes to slightly more than an hour.

The successful delivery of the injection does not establish that the experimental gene therapy works. Exin was still waiting to determine whether the intervention reduces seizures, and WIRED reported that the company was exploring additional AI models and more complex robotic procedures. The immediate result is therefore a demonstration of AI-assisted physical experimentation under controlled conditions, not proof that autonomous agents can safely replace trained researchers or surgeons.

## Sources
- [An AI Agent Injected Drugs Into a Live Mouse’s Brain](https://www.wired.com/story/an-ai-agent-did-brain-surgery-on-live-mouse/)