AI Agent Independently Selects and Injects Gene Therapy in Mouse Brain
A Claude model, without human decision-making, selected the drug, target region and injection coordinates and administered the treatment to an epileptic mouse. When a calibration error occurred, the agent refused to continue.
An AI agent independently performed a brain operation on an epileptic mouse. The system itself decided which drug would be used, which region of the brain would be treated, and at which coordinates the injection would take place. According to the lead researcher, Ocana-Santero, this is the first time an AI agent has independently selected a treatment and administered it to a living mammal.
The procedure initially proceeded as planned, until a calibration error occurred during the operation. The agent then stopped its work and could not be persuaded to continue. The episode shows that the system was not only able to make decisions, but also to set a boundary once the prerequisites were no longer met.
The case raises fundamental questions about how much decision-making latitude autonomous systems can be granted in medical procedures. So far, gene therapies and neurosurgical interventions in research have been strictly bound to human responsibility, with physicians and researchers monitoring and justifying every step. That software decided on the active agent, target site and dosage without human decision therefore marks a break with this practice, even though the experiment was limited to an animal.
The researchers point to the potential of such systems to make treatment processes more precise and faster. At the same time, the abort during the operation shows that automated procedures need reliable safety mechanisms. Whether and how the approach could be transferred to larger animal models or later to humans remains open according to the available information and would require further examination by independent bodies.
7
