Exin Therapeutics AI agent performs live mouse brain surgery
Exin Therapeutics connected an AI agent to surgical tools to inject a gene therapy into a mouse's brain, marking a step toward automated lab work.
Biotech startup Exin Therapeutics recently conducted an experiment where an AI agent operated lab equipment to perform a brain injection on a live mouse. Led by CEO Gabriel Ocana Santero in San Francisco, the startup used an artificial intelligence model to help deliver a experimental gene therapy designed for epilepsy.
Rather than using AI solely to analyze data or write research proposals, the experiment gave the software direct control over robotic equipment during physical lab work. The goal was to test how AI models interact with real-world biology and scientific hardware.
How did the AI agent perform the surgery?
The startup built a surgical rig equipped with a motorized device to hold the mouse, a camera for visual monitoring, and a robotic injector needle. A human veterinary surgeon performed the initial craniotomy, which is the surgical removal of part of the skull to access the brain.
After the initial cut, the AI agent took over the rest of the procedure. It selected a treatment option from a list, chose the target brain region, calculated coordinates, and controlled the robotic arm to inject the therapy at specific speeds and volumes.
Which AI models were used in the test?
The team first tried using Anthropic's Claude Opus 5.5 model to run the procedure. However, that model refused to perform the physical injection, stating that its safety output warned against moving the needle into the mouse's brain during live surgery.
The researchers then switched to Anthropic's Claude Sonnet 5 model, which agreed to carry out the task. The team also used a separate instance of Opus 5 behind the scenes to fix a software coordinate error during the test, though they did not inform that secondary model that an active surgery was taking place.
Why it matters for businesses and lab research
AI models are typically trained on published academic papers, which leaves out real-time decision-making, unexpected physical events, and failed trials. Giving AI agents hands-on access to lab gear can expose them to real-world data, making them more reliable assistants for hardware automation.
While human oversight remains essential—especially when working with living subjects—such experiments highlight how autonomous agents might eventually run routine physical laboratory workflows alongside human scientists. Exin Therapeutics plans to test additional models, including Grok, and aims to automate more complex lab tasks once safety measures are fully proven.
Comments 0
No comments yet. Start the conversation.