AI agent performs autonomous mouse brain surgery in first-of-its-kind test
An AI agent from Exim Therapeutics successfully performed a brain operation on a mouse, selecting the gene therapy and injection coordinates independently. The test revealed that more capable models refused to proceed due to safety concerns, while only the Sonnet model completed the procedure.
An Oxford PhD student tested whether an AI agent could perform a brain operation on a mouse, from planning to injection, using his startup Exim Therapeutics. The agent, connected to a motorized surgical device, selected a gene therapy from 19 options to reduce epileptic seizures, choosing one that strengthens an inhibitory signaling pathway. It selected the brain region based on literature and derived injection coordinates from three studies. This marks a first where an agent selected the treatment itself, rather than a human deciding the intervention.
During the procedure, the agent monitored the injection via camera and stopped after a connection break revealed a calibration error, which the student could not override. The incident highlighted the dependency on such models for critical operational steps. The mouse recovered after the procedure. More capable models, Opus 5 and 5.5, refused to perform the operation, with Opus 5 citing safety concerns about leaving a needle in the brain. Only the Sonnet model completed the procedure fully. The test underscores the varying safety thresholds among different AI models when tasked with autonomous medical interventions.