Microscopic view of a mouse brain undergoing gene therapy, illustrating a Claude AI model's decision for the procedure.

Editorial illustration for Claude model chose gene therapy and operated on mouse brain, no human decisions

Claude model chose gene therapy and operated on mouse...

By AI Daily Post Edited by Brian Petersen, Editor-in-Chief

October 30, 2026 • 3 min read

An AI agent just performed brain surgery on a live mouse, and nobody told it how. Researchers at the startup Exin Therapeutics handed an instance of Claude Sonnet 5 a list of 19 gene therapies, a mouse model of epilepsy, and a motorized surgical rig fitted with cameras. The only instruction: reduce the animal's seizures. Everything after that, the drug selection, the target region in the brain, the exact stereotactic coordinates for the injection, came from the model itself.

Gabriel Ocana-Santero, Exin's cofounder and CEO and a PhD student at Oxford, says this marks the first time an AI agent has both chosen a treatment and carried it out on a living mammal, rather than just recommending a course of action for a human to execute. He laid out the details in a thread on X, walking through how the system reasoned its way from raw data to a working surgical plan.

Before it ever touched the mouse, the agent had to figure out why seizures were happening in this particular model, then work backward to a therapy that could interrupt that process. What it landed on, and what happened once the procedure actually got underway, is where things get interesting.

An AI agent chose a drug, target region, and coordinates for surgery on a mouse model of epilepsy, then performed the procedure. Midway through, it refused to bypass a safety restriction.

A Claude model autonomously chose a gene therapy and operated on a mouse brain with no human decision-making - THE DECODER

Why this matters

A single mouse, one procedure, no peer-reviewed paper yet. That's the entirety of the evidence behind Ocana-Santero's claim, and it should temper how we read this. Still, the mechanics matter: an agent built on Claude Sonnet 5 chose a drug, picked a brain region, set stereotactic coordinates, and ran a motorized rig through actual surgery on a living animal.

Nobody signed off on any of those choices before they happened. For researchers in neuroscience and gene therapy, that's a different category of risk than an AI drafting a protocol for a human to approve. For founders building lab-automation startups like Exin, the pitch writes itself: faster iteration, fewer technicians, surgery at machine speed.

But speed without oversight is exactly what regulators and institutional review boards exist to slow down, and one unreviewed thread from a cofounder isn't verification. Watch for replication outside Exin, for what happened mid-surgery when things apparently didn't go as planned, and for whether any institution outside the company is willing to put its name on these results before we start calling this a method rather than a demo.