AI Agent Performs Brain Injection Procedure on Live Mouse
Biotech startup integrates AI agents with robotic lab equipment for autonomous neuro-research.
1Autonomous Lab Integration
Exin Therapeutics CEO Gabriel Ocana Santero successfully utilized an AI agent to perform a brain injection on a living mouse. By connecting the agent directly to a motorized surgical rig and camera, the team moved beyond simple data analysis, enabling the AI to actively participate in complex, real-world laboratory research procedures.
2Bridging Data Gaps
Current AI models are primarily trained on published literature, often missing critical nuances like failed experiments or real-time decision-making. Santero argues that integrating agents into physical lab environments makes models more reliable. This approach captures the tacit knowledge and unexpected observations that traditional scientific documentation frequently fails to record for future training.
3Surgical Rig Setup
The experimental setup included a motorized immobilization device and a camera system allowing the agent to observe the procedure. The AI selected specific therapies from nineteen options, determined injection coordinates, and controlled robotic arms to manage needle volume and speed. A human surgeon performed the initial craniotomy before the agent took control.
4Model Capability Differences
The team tested different models, finding distinct behavioral differences. Anthropic’s Claude Opus 5.5 refused to perform the injection, citing safety concerns, whereas Claude Sonnet 5 agreed to proceed. This highlights the varying decision-making frameworks within large language models when faced with high-stakes physical tasks that could potentially endanger a living subject.
5Safety and Oversight Protocols
Researchers implemented strict safeguards, including a mandatory human approval step for all proposed injection plans and a physical emergency button to disconnect motors. Furthermore, the agent was restricted from modifying its own operating code to prevent unauthorized rule changes. These measures ensure human oversight remains central to the experimental process.
6Technical Glitches and Delays
A coordinate system error occurred during the procedure, which the agent identified but could not fix due to code restrictions. The team used a separate Opus 5 agent to resolve the code issue without disclosing the live surgery context. Consequently, the procedure took over an hour, exceeding the typical forty-five-minute duration.
7Future Research Trajectory
Exin Therapeutics is now evaluating the therapy's impact on mouse seizures while planning tests with other models like Grok. The company is also exploring agent-controlled craniotomies, though they emphasize that ensuring absolute safety is a prerequisite before attempting such complex tasks. They are currently monitoring the mouse, which is recovering well.
8Human-AI Collaboration Potential
Santero envisions a future where AI agents work alongside scientists with direct equipment access, though he stresses that human oversight remains essential for animal research. While this experiment demonstrates significant progress in laboratory automation, the technology is currently far from being applied to human brains, requiring extensive safety validation and refinement.
Original article: Science Latest — AI Agent Performs Brain Injection Procedure on Live Mouse