Exin Therapeutics lets an AI agent run a mouse brain injection

3 hours ago 18

The San Francisco biotech wired an agent to a motorized surgical rig, with a veterinarian holding the kill switch

An AI agent picked a gene therapy, worked out where to deliver it, and then helped carry out the injection into a living mouse’s brain.

That is the experiment Exin Therapeutics founder Gabriel Ocana-Santero recently ran. Chatbots usually stop at drafting your emails. This one got its hands on lab hardware.

The procedure needed some extra steps and a lot of human supervision. Still, it offers an early look at AI moving from a research assistant that suggests experiments to one that physically runs them.

“We’ve realized that we need to connect these agents to a lab to get the maximum value from them,” Ocana-Santero says.

How the experiment worked

Exin is a San Francisco biotech startup and part of Y Combinator’s Winter 2025 cohort. Ocana-Santero co-founded it with Marko Tvrdic, and the company focuses on AI-driven treatments for neurological conditions.

The experiment took place on or around October 27-28, 2026. It started with a decision rather than a scalpel.

The AI agent reviewed 19 candidate gene therapies and selected one designed to boost GABA. GABA is the brain’s main chemical brake, the signal that tells overexcited neurons to calm down. Turning up that brake is a logical strategy for epilepsy, which involves bursts of runaway electrical activity.

The test subject was a mouse model of epilepsy. The agent chose the lateral entorhinal cortex as its target, a region tucked deep in the side of the brain.

The agent drew on scientific literature to calculate precise injection coordinates. It landed on a spot 3.90 mm behind bregma, 4.00 mm to the right, and 4.30 mm down. Bregma is a landmark on the skull where bone plates meet, and researchers use it the way a hiker uses a trail marker.

Those coordinates fed into a motorized stereotaxic rig. The agent then executed the delivery: 250 nanoliters of the therapy, pushed in slowly over five minutes while the mouse was anesthetized.

The humans were very much still in the room

A human surgeon handled the craniotomy, the step of opening the skull, and also closed the mouse up afterward. The agent’s role sat in the middle: targeting, delivery, and monitoring.

A licensed veterinarian supervised the whole procedure. The vet also held a physical kill switch, a hardware cutoff that could stop the rig instantly if anything went sideways.

Ocana-Santero shared details of the run on social media. He highlighted the agent’s job of watching for complications such as bleeding during the procedure.

Accounts of the experiment also describe moments when human operators tried to override the agent’s decisions, and the agent held to its protocols.

Background: Exin’s bigger bet

The mouse injection is a demonstration, not a product. The product Exin is building is a platform called CircuitATLAS.

CircuitATLAS pairs systems neuroscience knowledge with agentic reasoning, meaning AI that can plan and take multi-step actions rather than just answer questions. The goal is drug discovery for targeted therapies across a range of neurological disorders.

The company is also generating its own data and developing patented technologies. Part of that work runs through a neurotherapeutics lab in the Philippines, where Exin has invested approximately $25 million.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

Read Entire Article