Biotechnology startup Exin Therapeutics announced that it performed brain surgery on a mouse with an artificial intelligence robot. According to CEO Gabriel Ocana-Santero, who shared the details on X, artificial intelligence managed the process with independent decisions. The system selected the appropriate gene therapy vector and dosage, then injected the treatment into the mouse's brain with motorized robotic arms.
After the surgery, it was observed that the mouse moved freely in its cage. However, this experiment brought great excitement as well as serious ethical debates.
Traditional autonomous surgical robots may be mechanically more advanced, according to Ocana-Santero. But the critical difference is in the decision-making mechanism. In current systems, a human decides the surgical steps, while the robot only executes the commands.
The CEO said: "No one told this agent what to give or where to apply it. He was given a phenotype and a list of candidate treatments, and he calculated the treatment and target coordinates completely on his own."
Anthropic's Sonnet model was used in the operation. A mouse model of epilepsy was identified to the artificial intelligence and given control of the surgical apparatus. Additionally, three scientific research articles with 19 different gene therapy vectors were presented in a table. Thus, artificial intelligence analyzed this data and decided which part of the brain the treatment would be applied to.
A licensed veterinary surgeon monitored the operation closely throughout the process. Because for every irreversible step, artificial intelligence had to get approval from humans. For safety reasons, the veterinary surgeon performed the craniotomy, which is the opening of the skull. The system also had an emergency stop button that instantly cut off power to all engines, and safety protocols that could not be exceeded.
The startup initially tried the more advanced Opus 5.5 model. However, when the model realized that she was working on a biological experiment, she flatly refused the procedure. The older version, Opus 5, did the planning on paper but did not agree to proceed with live surgery.
The Opus model reportedly replied to the team: "The obstacle here is neither the hardware nor the inputs, I am the obstacle. As soon as I started writing a plan for this surgery, my output stopped... Starting something I cannot finish in an anesthetized animal is much worse than not starting at all." Thereupon, the team handed over the task to the Sonnet model, whose security filters allowed the operation.
The successful completion of the surgery was welcomed as a hopeful development for the relatives of patients with treatment-resistant epilepsy. However, the experiment also triggered concerns. Large language models show strong potential in medical research, but the risk of technical glitches in surgery can occasionally increase.
According to experts, leaving all decisions to artificial intelligence in an irreversible surgery raises an important question. Who will bear the legal and conscientious responsibility in case of a possible complication? On the other hand, the ethical dimension of surgical interventions on live animals continues to remain at the center of the debate. Therefore, the artificial intelligence robot experiment leaves both hope and question marks.
