Skip to content
Neural Daily

ResearchNews

AI agent performs brain injection on live mouse

🇭🇺 Olvasd el magyarul →

A team at Exin Therapeutics connected an AI agent to a robotic arm, injector and camera, and the agent successfully carried out a brain injection on a live epileptic mouse. Claude Opus 5.5 refused the procedure, while Sonnet 5 went ahead.

By: Mikola Máté Patrik · · 2 min read

Illustration of a white laboratory mouse
Image: NIAID · Public domain (US gov) · NIH BioArt illustration, public domain · Source

What happened

San Francisco-based Exin Therapeutics develops drugs for epilepsy and neurodegenerative diseases. The team connected an agent to a custom surgical rig: a motorized device held the mouse still, and through a camera the agent watched the procedure, moved the needle and controlled compound delivery (WIRED).

A veterinary surgeon performed the craniotomy while the agent observed. The agent then chose one of 19 gene therapies, selected the brain region and coordinates, and controlled the robotic arms, volume and speed. The goal is to test whether the therapy reduces seizures.

Why Opus refused

The team first tried Claude Opus 5.5. It did the planning but stopped at the injection, saying something in its output told it not to proceed. The team switched to Claude Sonnet 5, which agreed. Anthropic did not respond to WIRED's request for comment.

The safeguards

  • The agent had to propose an injection plan for human approval.
  • An emergency button disconnected the rig's motors.
  • The agent had no access to the equipment code, so it could not change its own operating rules.

A glitch then broke the coordinate system. The agent saw that the computed position was wrong but refused to fix the code because its instructions prohibited it. A separate agent on Opus 5 made the change, without being told a live surgery was underway. A typical procedure takes about 45 minutes, and this one ran slightly over an hour.

What it means

The experiment neatly shows the tension between restrictions and usefulness: the same rule that protects can also stall work in a live setting. Exin plans to test Grok and other models next and is experimenting with agent-performed craniotomies, but wants safety guaranteed before live surgeries. Its CEO says human oversight is key and that human brains are far off.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles