An AI agent injected gene therapy into a live mouse brain

Exin Therapeutics gave an AI agent a task usually kept firmly in human hands: help inject a gene therapy into the brain of a living mouse with epilepsy. WIRED reports that a veterinary surgeon first opened the skull. The agent then chose one of 19 therapy options, selected a brain region and coordinates, and controlled the robotic arms and injector.

The experiment was not a clean demonstration of autonomy. The first model the team tried refused to guide the needle, warning that it could stop midway and endanger the animal. The team switched models. During the procedure, a coordinate-system glitch produced a wrong position. The agent noticed the error but would not change the underlying code because its instructions prohibited that. A second agent corrected the code, and a human remained able to stop the rig.

The procedure took a little over an hour, compared with about 45 minutes for a typical injection, according to the founder. The mouse was reported to be recovering well. The important result is still pending: the team has not yet assessed whether the gene therapy reduces seizures.

That gap matters. The story is striking because an AI system crossed from planning into physical laboratory work, not because it has proved a treatment works. The safeguards, refusal and software glitch are part of the experiment too. They show why a capable agent around living subjects still needs a human operator, a reliable emergency stop and a clear record of what it actually accomplished. WIRED's report provides the full account.