Artificial Intelligence Autonomous Mouse Brain Surgery Completed by Exin Therapeutics
An artificial intelligence agent independently selected a gene therapy treatment, calculated brain coordinates, and operated on a live mouse to treat epilepsy, 3DNews reported. Gabriel Ocana-Santero, chief executive officer of biotechnology startup Exin Therapeutics, announced the milestone on social media. While autonomous surgical robots exist in laboratory and clinical settings, previous systems relied on human operators to decide the treatment plan and surgical site. In this experiment, the software developed its own therapeutic strategy without human direction on which drug to use or where to inject it, according to 3DNews.
Anthropic Claude Model Controls Robotic Stereotaxic Setup
The procedure took place during an experiment involving a mouse induced with epilepsy. Researchers provided the AI agent, operating on Anthropic’s Claude Sonnet model, with a spreadsheet containing 19 potential gene therapy vectors and access to three scientific publications, as reported by 3DNews and iXBT.com. The model analyzed the data, selected a therapeutic approach, and determined the injection site. According to details shared by El.kz, the system calculated coordinates of 3.90 mm behind the bregma, 4.00 mm to the right, and 4.30 mm deep. The AI then commanded a motorized stereotaxic surgical installation to deliver 250 nanolitres of the substance over five minutes.
AI Agent Reportedly Controlled Mouse Brain Injection After Another Claude Model RefusedA biotech startup says it used an AI agent to control a gene therapy injection into a living mouse's brain. But the most important part of the experiment may be what the AI was allowed to… pic.twitter.com/TY2VEQvuMB
Human Supervision and Model Refusals During the Procedure
The operation was not fully autonomous, as safety protocols required human intervention at critical stages. A licensed veterinary surgeon performed the craniotomy to open the mouse skull because the task is not yet trusted to an AI agent, as noted by 3DNews. The veterinarian also monitored the procedure with an emergency stop button and approved irreversible steps before the software continued. Technical hurdles also emerged during testing. Gabriel Ocana-Santero stated that earlier attempts to use Anthropic’s Opus 5.5 model failed when the system refused to participate in a biological experiment. An earlier Opus 5 version drafted a plan on paper but declined to execute it on a live animal due to safety boundaries.
Uncertain Treatment Efficacy and Safety Questions in Robotic Procedures
Exin Therapeutics reported that the mouse recovered well from the intervention and remains active. However, iXBT.com noted that the experiment does not yet confirm whether the chosen gene therapy successfully reduced the frequency of epileptic seizures, as the company has not yet published detailed protocols, measurements, or final trial results. The demonstration has also renewed broader debates regarding accountability and safety when artificial intelligence assumes direct control over invasive medical procedures. Researchers and industry observers point to previous data showing that introducing AI decision-making into sinus surgeries led to a significant increase in technical errors, underscoring the challenges of integrating autonomous systems into clinical environments.
Frequently Asked Questions About the AI Brain Surgery Experiment
What specific drug and delivery method did the AI select?
The AI agent chose from a provided table of 19 gene therapy vectors designed to deliver viral genomes. It selected a specific vector and administered 250 nanolitres of the substance directly into the target brain structure over a five-minute period.
How did the AI determine the exact injection coordinates?
The model calculated the target coordinates by analyzing three scientific publications provided by the researchers. It averaged the reported values to determine a location 3.90 mm behind the bregma, 4.00 mm to the right, and 4.30 mm deep.
What safety mechanisms protected the animal during the procedure?
A licensed veterinarian performed the skull opening and supervised the operation with a physical emergency stop button capable of cutting power to all robotic drives. Furthermore, the AI required explicit human approval before executing any irreversible step.
Keep reading