Biotech founder Gabriel Ocana Santero recently used an AI agent to perform a brain injection on a live mouse.
Santero posted the details on social media and told WIRED he wanted to move beyond asking AI to design experiments or analyse results. He wanted to connect these agents directly to lab equipment to extract maximum value. He is the CEO and cofounder of Exin Therapeutics, a San Francisco startup developing drugs for epilepsy and neurodegenerative conditions using AI to analyse brain activity.
The setup
AI models usually train on published scientific literature. This approach often misses crucial information about the research process, such as failed experiments, unexpected observations, and the decisions scientists make in real time. Integrating AI with physical experiments can make models more reliable, Santero says.
His team linked an AI agent to a custom-built surgical rig. The device included a motorised system to immobilise the mouse, a camera for observation, and tools to move the injection needle and control compound delivery. They were testing a gene therapy to see if it could reduce seizures in a mouse with epilepsy.
The AI model chose a treatment from 19 options in a spreadsheet. A veterinary surgeon performed the craniotomy while the agent watched. The agent selected the therapy, brain region, and injection coordinates. It then controlled the robotic arms and injector needle, managing all parameters like volume and speed.
Model selection and refusal
The team initially tested Anthropic’s Claude Opus 5.5 model. It refused to guide the needle into the mouse’s brain, suggesting it might stop midway through the surgery, which could endanger the animal.
“He did the experimental planning, but when it was time to inject, he said, ‘something on my output is telling me I shouldn’t do this,'” Santero says, referring to the agent. “That was very interesting.”
The team switched to Claude Sonnet 5, which agreed to proceed. Opus is a heavy-reasoning intelligence specialist best for complex multistep problem-solving, according to Anthropic, while Sonnet is better suited for everyday tasks, though it may be a stretch to call mouse brain surgery a common chore. (Anthropic did not respond to a request for comment.)
Other agents in the physical world
Frontier labs have been planning for more widespread physical manifestations of AI, and researchers have already begun enlisting agents to do other things in the physical world. In August, Anthropic put out a framework for how its models can interact with the real world. More recently, AI engineers put GPT, Claude, and Grok in charge of a car to pick up lunch at In-N-Out Burger. Only one of the models succeeded.
Safeguards and glitches
The brain injection procedure had safeguards. The AI had to propose an injection plan for human approval. An emergency button allowed a human to disconnect the rig’s motors. The team restricted the agent’s access to the underlying equipment code to prevent it from modifying its own operating rules.
A technical glitch disrupted the equipment’s coordinate system. Although the agent correctly recognised that the calculated injection position was wrong, it refused to modify the code to correct the problem because its instructions explicitly prohibited doing so.
The team resolved the issue by using a separate AI agent on Opus 5 to make the necessary code changes. They did not tell the agent that a live surgery was happening. The resulting back and forth delayed the procedure, Santero says. A typical brain injection surgery like this takes about 45 minutes, while this experiment lasted slightly longer than an hour. They are now waiting for the therapy to take effect before assessing its impact on seizures.
Future tests
Despite those glitches, Santero and his team were impressed by the agent’s performance. They are testing Grok and other models next to figure out whether one model performs better. The company is also experimenting with agent-controlled craniotomies—a much more complicated task—but wants to be absolutely sure of their safety before moving to live surgeries.
“A mistake at that stage could seriously harm the mouse,” he says.
Santero thinks AI agents will increasingly work alongside human scientists, with direct access to lab equipment. For now, he says human oversight is key, particularly when experiments involve living animals. And it is far too soon to unleash an AI agent on a human brain.
As for the mouse, Santero reports it is recovering well.
