Exin Therapeutics co-founder and CEO Gabriel Ocana-Santero said in an October 28, 2026, post that an AI agent selected a gene-therapy candidate and brain target for a mouse with epilepsy, then controlled a motorized rig to deliver an injection. A veterinary surgeon had opened the skull, and people approved the injection plan and held an emergency stop.
What the agent reportedly did
Ocana-Santero’s post described 19 gene-therapy candidates and three research papers used to derive injection coordinates. He said the agent selected a treatment and target, then controlled the rig’s arms and injector to deliver 250 nanoliters while monitoring the entry site with a camera.
Where people stayed involved
A veterinary surgeon performed the craniotomy, the operation that opens the skull. People approved the proposed injection plan, and a human-held emergency button could disconnect the rig’s motors. The account describes a division of work: the agent handled treatment selection and the injection task, while people performed the skull-opening step and retained approval and emergency-stop roles.
Model refusals, a coordinate glitch and the outcome
WIRED reported that Claude Opus 5.5 declined to guide the needle; Claude Sonnet 5 proceeded with the injection. After a coordinate-system glitch led the agent to recognize an incorrect position, it refused to change code that its instructions barred it from editing. A separate Claude Opus 5 agent made the code changes without being told a live procedure was underway. The procedure took slightly more than an hour, compared with about 45 minutes for a typical injection of this kind, according to Ocana-Santero’s account to WIRED.
Ocana-Santero said the mouse was recovering well. In its October 31, 2026, account, WIRED said the team was still assessing whether the treatment reduced the mouse’s seizures.
