A Milestone in Automated Medicine

In a development blurring the lines between science fiction and reality, an artificial intelligence agent has successfully executed critical phases of a brain operation on an animal model. The experiment, detailed by Oxford-trained researcher and Exin Therapeutics co-founder Gabriel Ocana-Santero, demonstrates the growing capability of AI to make complex clinical decisions and control surgical hardware without direct human instruction on treatment methodology.

The procedure, conducted on a mouse model of epilepsy, involved connecting an advanced AI model to a custom-built, motorized surgical rig. Rather than following a rigid, pre-programmed script, the agent was presented with a spreadsheet containing 19 potential gene therapies and left to determine the optimal course of action independently.

Autonomous Decision-Making and Precision Execution

Tasked with treating the neurological condition, the AI agent analyzed scientific literature to determine the appropriate therapeutic compound, identified the precise region of the brain to target, and calculated exact stereotaxic coordinates. It then assumed direct control of the robotic rig to administer the injection.

Throughout the procedure, the agent monitored its progress via an integrated camera feed, actively scanning for complications such as bleeding or fluid backflow at the surgical site. Operational readouts logged by the system during the procedure highlighted a high degree of situational awareness, noting precise volumetric measurements and visual confirmation of suture placement as the incision was closed.

Safety Protocols and Model Limitations

The experiment also highlighted the critical role of hardcoded safety boundaries in autonomous medical systems. When a manual adjustment during a pause desynchronized the physical rig from its digital coordinate reference point, the AI detected the anomaly. Encountering a calculated Y-axis target exceeding the hardware's operational range, the agent repeatedly refused to proceed, prioritizing safety over task completion and forcing human operators to intervene and recalibrate the system.

Furthermore, the experiment underscored disparities in how different foundational models approach high-stakes physical interventions. While the primary model successfully completed the live procedure under strict human supervision—including a licensed veterinarian holding a physical emergency kill switch—other models, such as Anthropic's Opus 5.5, declined to participate, citing the ethical and physical risks associated with biological experimentation.

While the mouse successfully recovered from the procedure, researchers note that determining the long-term efficacy of the AI-selected gene therapy in reducing epileptic seizures will require ongoing observation. Nevertheless, the experiment marks a profound leap forward in the potential application of autonomous systems within neurosurgery.