AI agent performs autonomous brain surgery on mouse, selecting drug and target independently

Researchers at Exin Therapeutics, led by CEO Gabriel Ocana-Santero, connected an AI agent to lab equipment to independently select a gene therapy and perform brain injection surgery on a live mouse model of epilepsy. While Anthropic's Claude Opus models refused to perform the injection due to safety safeguards, Claude Sonnet 5 executed the procedure. Despite coordinate glitches that extended surgery past one hour, human oversight and an emergency stop ensured safety, and the mouse recovered well.

AI-controlled mouse brain surgery

• ▪The custom surgical setup built for Exin Therapeutics' mouse epilepsy experiment included a motorized device to immobilize the mouse, a camera for AI observation, and a needle injector
• ▪Gabriel Ocana-Santero and researchers at Exin Therapeutics used an AI agent to autonomously select a gene therapy and carry out a brain injection procedure on a live mouse model of epilepsy

Model safety refusals

• ▪Exin Therapeutics switched to Claude Sonnet 5, which was the only tested model willing to carry out the live mouse epilepsy brain surgery from start to finish
• ▪Anthropic's Claude Opus 5.5 model refused to guide the injector needle into the live mouse's brain, citing risks that stopping mid-procedure could endanger the animal

AI lab automation context

• ▪Exin Therapeutics is experimenting with agent-controlled craniotomies, but is delaying live animal tests until safety protocols can be fully assured
• ▪Gabriel Ocana-Santero is CEO and cofounder of Exin Therapeutics, a San Francisco-based startup developing drugs for epilepsy and neurodegenerative diseases by analyzing brain activity