Exin Therapeutics successfully utilized a Claude Sonnet 5 model to autonomously select a gene therapy and execute a surgical procedure on a living mouse. Gabriel Ocana-Santero, cofounder and CEO of Exin Therapeutics and a PhD student at the University of Oxford, reported that the AI agent analyzed scientific literature, picked a specific drug and target brain region, calculated injection coordinates, and controlled a motorized surgical rig to treat a mouse model of epilepsy.
During the procedure, the agent monitored the surgical field through cameras, made numerous observations, and successfully tracked the injection process. However, mid-surgery safeguards were triggered after a connection dropped, causing the agent to detect a position outside its physical range of motion and refuse to continue. Although the researchers attempted to persuade it to proceed, the model correctly identified a system error caused by manual adjustments, demonstrating both the reliability of its safety protocols and the challenges of human-AI alignment during time-sensitive tasks.
This event marks the first recorded instance of an AI agent independently choosing a medical treatment and delivering it to a living mammal. Why it matters: The milestone highlights the advancing capabilities of artificial intelligence in autonomous medical decision-making while underscoring critical questions regarding control, alignment, and reliability in clinical applications.