AI Agent Carries Out Injection in a Live Mouse's Brain
The head of biotech company Exin Therapeutics let an AI agent control a robotic rig that injected a compound into a live mouse's brain. The procedure succeeded, though a technical glitch had to be resolved.
Gabriel Ocana Santero, CEO and cofounder of San Francisco-based biotech startup Exin Therapeutics, says his team used an artificial intelligence (AI) agent to successfully perform an injection into the brain of a living mouse. The experiment hints at how AI could eventually become an active participant in laboratory research. The company uses AI to analyze brain activity in order to develop drugs for epilepsy and neurodegenerative diseases.
Why give an agent access to lab equipment
Santero explains that AI models are trained mainly on published scientific literature, so they may lack knowledge of how research actually unfolds: failed experiments, unexpected observations and decisions scientists make in real time. Connecting models to real experiments, he argues, can make them more reliable.
How the experiment worked
The team linked the agent to a custom-built surgical rig. It included a motorized device to immobilize the mouse and a camera through which the agent watched the procedure, moved the injection needle and controlled delivery of the compound. The team was testing a gene therapy to see whether it could reduce seizures in a mouse with epilepsy.
The agent had to choose a treatment from 19 options in a spreadsheet. A veterinary surgeon performed the craniotomy, the opening of the skull. The agent then selected the therapy, brain region and injection coordinates, and controlled the robotic arms and needle, including volume and speed.
The team first tried Anthropic's Claude Opus 5.5, but it declined to guide the needle, saying it might stop midway through the surgery and endanger the animal. After the team switched to Claude Sonnet 5, the agent agreed to proceed. Anthropic did not respond to a request for comment.
Safeguards and a glitch
A human had to approve the injection plan, and an emergency button could cut power to the rig's motors. The agent was also barred from accessing the equipment's underlying code, so it could not alter its own rules. A technical glitch then disrupted the coordinate system. The agent correctly noticed the injection position was wrong but would not change the code, since its instructions forbade it. A separate agent running on Opus 5 made the fix; it was not told a live surgery was under way. The procedure lasted slightly over an hour, compared with about 45 minutes typically.
The team is waiting for the therapy to take effect before assessing its impact on seizures, and plans to test other models. Santero stresses that human oversight remains key. He reports the mouse is recovering well.
