Updated
Updated · WIRED · Oct 31
AI Agent Injects Gene Therapy into Live Mouse Brain in 1-Hour Surgery
Updated
Updated · WIRED · Oct 31

AI Agent Injects Gene Therapy into Live Mouse Brain in 1-Hour Surgery

3 articles · Updated · WIRED · Oct 31

Summary

  • A custom AI agent completed a brain injection on a live epileptic mouse, selecting one of 19 gene-therapy options, choosing the target brain region and coordinates, and controlling the robotic needle and delivery settings.
  • The test was designed to see whether giving AI direct control of lab equipment can improve models trained mostly on published papers, which often miss failed experiments and real-time lab decisions.
  • Claude Opus 5.5 planned the procedure but refused to guide the needle into the brain, so the team switched to Claude Sonnet 5, while a veterinary surgeon had already performed the craniotomy.
  • A coordinate-system glitch extended the surgery from about 45 minutes to just over 1 hour after the agent spotted the error but would not alter blocked code; a separate Opus 5 agent made the fix under human safeguards.
  • Exin Therapeutics is still waiting to see whether the therapy reduces seizures, plans to compare Grok and other models next, and says human oversight remains essential for animal experiments.

Insights

Why did one Claude model refuse a live mouse brain injection while another completed it, and what does that reveal about AI lab safety?
Can an AI-guided mouse brain surgery that needed human safeguards and bug fixes really mark the start of self-driving biology?
If AI can already control needles in live-animal experiments, what must happen before science trusts it with more complex procedures?