Claude model autonomously performed brain surgery on a mouse

Researchers used a language model from Anthropic to perform a surgical operation on a living organism without human intervention. The experiment was conducted on a mouse with symptoms of epilepsy.

During the procedure, the AI-based algorithm independently decided on the choice of gene therapy, identified a specific target in the brain tissue, and calculated the precise coordinates for the drug injection. According to the research team leader, Ocana-Santero, this is the first recorded instance of an autonomous agent fully planning and executing a medical intervention in a living subject.

The process encountered technical difficulties during operation: due to a calibration error in the middle of the surgery, the system refused to continue the procedure. The scientists were unable to persuade the agent or resume the task.

The results demonstrate both the new possibilities for automating complex medical procedures and the critical limitations of modern autonomous systems in unforeseen situations.