A Claude Sonnet 5 agent picked a gene therapy and carried out brain surgery on a mouse, and stopped when it found a calibration error
Exin Therapeutics says this is the first time an AI agent has both chosen a treatment and delivered it to a living mammal.
An AI agent running on Claude Sonnet 5 chose a treatment for an epileptic mouse and performed the surgery to deliver it, according to Gabriel Ocana-Santero of the startup Exin Therapeutics, as reported by The Decoder. The team gave the agent a list of 19 possible gene therapies, a mouse model of epilepsy in which researchers induce seizures to test treatments, and access to a custom motorized surgical rig with cameras, a microscope and a syringe. Its only instruction was to reduce the mouse's seizures. The agent selected the drug, the brain region and the stereotactic coordinates on its own, then ran the procedure through the rig. Partway through, it detected a calibration error and refused to continue past its safety limits. Ocana-Santero says this is the first time an AI agent has both chosen a treatment and delivered it to a living mammal. The report does not include independent verification or treatment outcomes. The demonstration matters for both laboratory automation and oversight. It shows a general-purpose model making consequential biomedical decisions and acting physically on a living animal, and it shows a built-in safety constraint stopping the procedure.