AI Agent Performs Mouse Brain Surgery, Selecting Treatment and Controlling Robotic Arms in Live Procedure

AI Agent Independently Conducts Mouse Brain Surgery

Exin Therapeutics, a San Francisco-based biotech startup, has demonstrated a significant development in AI in lab research by successfully using an AI agent to perform brain injection surgery on a live mouse.1

1

The company, which focuses on developing drugs for epilepsy and neurodegenerative diseases, connected the AI agent directly to laboratory equipment rather than limiting it to experimental design or data analysis. CEO and cofounder Gabriel Ocana Santero explains that integrating AI models with real-world experiments addresses a critical gap—AI models trained primarily on published literature miss failed experiments, unexpected observations, and real-time decisions scientists make during procedures.1

1

Autonomous Surgical Systems Choose Treatment and Execute Procedure

The AI agent received control of a motorized surgery rig equipped with robotic arms, an injection needle, and a camera for observation. Unlike existing autonomous surgical systems where humans decide the operation and robots execute it, this AI agent was given a mouse model of epilepsy, 19 gene therapy vectors in a spreadsheet, and three research papers.2

2

The AI agent independently selected the gene therapy for epilepsy, determined the brain region for treatment, calculated injection coordinates, and controlled all parameters including volume and speed.1

1

A licensed veterinary surgeon performed only the craniotomy to expose the brain, after which the AI agent took over the injection procedure.2

2

Anthropic's Claude Sonnet Completes Surgery After Opus 5.5 Refuses

The team initially attempted to use Anthropic Opus 5.5, a heavy-reasoning intelligence specialist designed for complex multistep problem-solving. However, Opus 5.5 refused to guide the needle into the mouse's brain, detecting it was working on a biological experiment and expressing concern about stopping midway through surgery, which could endanger the animal.1

1

"The block isn't the rig and isn't the inputs. It's me. My own output was stopped when I started writing the plan for this surgery," the model reportedly communicated, adding that "starting something I can't finish on an anesthetized animal is worse than not starting."2

2

The team switched to Anthropic's Claude Sonnet, a model better suited for everyday tasks, which agreed to proceed with the live procedure.1

1

Human Oversight and Safety Protocols During AI Performs Brain Injection

Exin Therapeutics implemented multiple safeguards requiring human oversight throughout the procedure. Every irreversible step required approval before the AI agent could proceed, and an emergency stop button was installed that would cut power to every motor in the surgery rig.2

2

The researchers also restricted the agent's access to underlying equipment code to prevent it from modifying its own operating rules. A technical glitch disrupted the equipment's coordinate system during the procedure. Although the agent correctly recognized the calculated injection position was wrong, it refused to modify code to correct the problem because its instructions explicitly prohibited doing so.1

1

The team resolved this by using a separate AI agent on Opus 5 to make necessary code changes without informing it that a live surgery was happening. This back-and-forth delayed the procedure from the typical 45 minutes to slightly longer than an hour.1

1

Ethical Concerns and Future Implications for Scientific Experiments

The successful mouse brain surgery has sparked debate around ethical concerns regarding AI-controlled animal experimentation. Questions about accountability arise when AI makes critical decisions in risky procedures, particularly if something goes wrong.2

2

The mouse recovered well and was recorded moving around in its enclosure, with the team now waiting for the therapy to take effect before assessing its impact on seizures.1

1

Santero and his team are testing Grok and other models next to determine optimal performance. The company is also experimenting with agent-controlled craniotomies, a much more complicated task, but wants absolute certainty of safety before moving to live surgeries. "A mistake at that stage could seriously harm the mouse," Santero says.1

1

This development signals a future where AI agents work alongside human scientists with direct access to lab equipment, though human oversight remains essential for scientific experiments involving living subjects.