What changed
Exin Therapeutics, a biotech startup, claims it successfully performed brain surgery on a mouse using an AI agent. The AI agent selected a gene therapy vector, dosage, and brain region for treatment, then guided robotic arms to inject the medication. The procedure was conducted using Anthropic’s Sonnet model, after the Opus 5.5 model refused to proceed with a biological experiment.
What this means in practice
The AI agent was given a mouse model of epilepsy, a motorized surgery rig, 19 gene therapy vectors, and three research papers. It autonomously determined the treatment and coordinates without human intervention, though a licensed veterinarian oversaw the operation and approved irreversible steps. The AI also had an emergency stop button to halt the procedure. The team noted that the AI was not allowed to perform the craniotomy, a task reserved for the human surgeon. Safeguards prevented the AI from overriding certain decisions, requiring manual code adjustments during the operation.
Limitations and availability
The AI agent used was Anthropic’s mid-range Sonnet model, not the Opus 5.5, which refused to conduct the experiment. The procedure was conducted under strict oversight, with human involvement in critical steps. The startup emphasized that the AI did not make decisions about the craniotomy or irreversible actions, which required human approval. The experiment raises ethical concerns about AI autonomy in medical procedures and the use of animals in research. The supplied announcement does not specify whether the AI agent’s decisions were independently verified or how the success of the operation was measured.
