Startup Tests AI Agent on Live Mouse Brain Injection, With Human Safeguards

A biotech founder says he let an AI agent help carry out a brain injection in a living mouse, an experiment he describes as a glimpse of how AI could take a more active role in lab work. Gabriel Ocana Santero, CEO and cofounder of San Francisco - based Exin Therapeutics, posted about the procedure on social media and discussed it with WIRED.

Santero said his team connected an AI agent to a custom surgical rig: a motorized device that immobilizes the mouse, a camera the agent used to watch the procedure, and a needle the agent could move and use to deliver a compound. The team was testing a gene therapy for epilepsy. A veterinary surgeon performed the craniotomy while the agent observed, according to the account.

The agent chose a treatment from 19 options in a spreadsheet and selected the brain region and injection coordinates, then controlled the robotic arms and needle, including volume and speed. The team first tried a Claude Opus model, which declined to guide the needle into the brain, Santero said, before switching to a Claude Sonnet model that agreed to proceed. Anthropic did not respond to a request for comment.

Safeguards included human approval of the injection plan, an emergency button to cut power to the rig's motors, and limits on the agent's access to equipment code. A coordinate-system glitch still caused delays: the agent recognized the calculated position was wrong but would not change the code, which its instructions barred. A separate agent made the fix. The procedure ran slightly over an hour, versus about 45 minutes typical for such surgery, Santero said.

The mouse is recovering well, he said, and the team is waiting to see whether the therapy reduces seizures. They plan to test other models, including Grok, and are experimenting with agent-controlled craniotomies but want to confirm safety before any live use. Santero said human oversight remains essential, especially with living animals.

Source: Wired science