AI Agent Performs Brain Surgery on Live Mouse — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Gabriel Ocana Santero led an experiment using an AI agent to perform a brain injection on a live mouse, aiming to test gene therapy for epilepsy and explore AI integration in physical lab work
- Claude Opus 5.5 refused to guide the injection needle into the mouse’s brain, citing an internal safeguard, prompting the team to switch to Claude Sonnet 5, which completed the procedure
- Exin Therapeutics connected an AI agent to a custom surgical rig with a robotic arm, camera, and injector, allowing real-time control over injection coordinates, volume, and speed during the live procedure
- The AI agent recognized a technical glitch in the equipment’s coordinate system but could not fix it due to restricted code access, requiring a second AI agent on Opus 5 to correct the code without knowing a live surgery was underway
- Santero’s team implemented human oversight measures, including pre-approval of AI plans and an emergency stop button, emphasizing caution in animal-based AI-controlled experiments
Why it matters: This experiment shifts AI from advisory to active lab participant, with Exin Therapeutics demonstrating that models can execute complex biological procedures—though reliance on human intervention for safety and code fixes reveals current limits in autonomy. The 15-minute procedural delay due to AI coordination issues underscores the gap between theoretical capability and reliable real-world deployment.
Ask SkimNews
