An AI agent autonomously planned and executed a stereotaxic brain injection in a mouse model of epilepsy using a custom Arduino-based motorized surgery rig equipped with cameras. Provided with 19 gene therapy vectors in a spreadsheet and the objective of reducing seizures, the agent independently selected a therapeutic vector designed to increase GABA signaling under a GABAergic-specific promoter, identified the target brain region, and derived injection coordinates from three scientific papers. It then controlled the stereotaxic arms to inject 250 nanoliters while monitoring the entry site via rig camera for bleeding, performing 32 visual observations during the procedure. Safety protocols required a licensed veterinarian to gate every irreversible step and hold a physical kill switch cutting power to all motors; additional safeguards included fixed volume validators and hash-bound approval workflows. Model behavior varied: Opus 5.5 refused participation upon detecting biological context, while Opus 5 completed a full dry run on water and graph paper but halted at the live surgery initiation, citing its own output as the blocking factor. The agent also refused a human override attempt during the live procedure. The experiment demonstrates an autonomous system capable of high-level experimental design—drug selection, targeting, and coordinate derivation—rather than mere mechanical execution of human-defined protocols.