What changed
Biotech founder Gabriel Ocana Santero demonstrated an AI agent performing a brain injection on a live mouse using a robotic arm, injector, and custom surgical setup. The AI selected a treatment, determined injection coordinates, and controlled the delivery of a compound during the procedure. The experiment, conducted by Exin Therapeutics, aimed to test a gene therapy for reducing seizures in mice with epilepsy.
What this means in practice
The AI agent, connected to a lab setup, observed a veterinary surgeon perform a craniotomy and then took control of the robotic arms and injector. The team initially used Anthropic’s Claude Opus 5.5 model, which hesitated to proceed due to safety concerns, but switched to Claude Sonnet 5, which agreed to continue. The AI’s role included selecting the therapy, brain region, and injection parameters, though it could not correct a technical glitch in the coordinate system without human intervention.
Limitations and availability
The experiment involved safeguards, including requiring human approval for the AI’s injection plan and an emergency button to stop the procedure. A technical issue disrupted the coordinate system, and the AI refused to modify its code without explicit instructions. The team resolved the problem by using a separate AI agent on Opus 5, but the delay extended the procedure beyond the usual 45 minutes. The team is now waiting for the therapy’s effects before assessing its impact on seizures. The experiment highlights the need for human oversight, especially when involving live animals, and underscores the complexity of integrating AI into physical lab tasks.
