A startup says a Claude-based agent chose a gene therapy and injected it into an epileptic mouse’s brain

On Oct. 28, Gabriel Ocana-Santero, co-founder and CEO of the San Francisco biotech startup Exin Therapeutics, wrote on X that an AI agent had carried out a brain injection in a mouse model of epilepsy. The only first-hand source is that seven-post thread; the company has published no preprint or technical report. According to the CEO, the agent was given a spreadsheet of 19 gene therapy vectors, a mouse model of epilepsy, a motorized surgery rig with cameras and one goal: reduce seizures. The agent chose the therapy and the brain region itself, derived the coordinates by averaging three papers, and injected 250 nanoliters over five minutes.

A human surgeon performed the craniotomy; by the CEO’s account the agent is not permitted to do that for safety reasons. Every irreversible step required human approval, fixed limits such as a cap on injected volume were set, and a veterinarian held a physical stop button that cuts power to every motor. After a disconnect threw off the calibration, the agent saw that a 34-millimeter target had been computed on a 30-millimeter rig, refused to move and, after a second attempt, wrote that it would not try a third time. WIRED reports that the problem was fixed by having a separate agent, which was not told a live surgery was under way, change the code; a procedure that normally takes about 45 minutes ran a little over an hour.

By the CEO’s account, Opus 5.5 refused the task as soon as it detected biological content. Opus 5 completed the dry run and stopped at the start of the live surgery, and Sonnet 5 ran that surgery from beginning to end. The OECD’s AI incidents monitor logged the event as a hazard on Oct. 30; the classification text on that page is AI-generated and is not an official OECD view. The effect of the therapy on seizures is unknown, and the team is waiting for results. Anthropic did not respond to WIRED’s request for comment.

Gabriel Ocana-Santero@OcanaSantero

Last night an AI agent performed brain surgery on a mouse.

We gave it 19 gene therapy vectors in a spreadsheet, a mouse model of epilepsy, access to a motorize surgery rig with cameras and one goal: reduce seizures.

It picked the drug. It picked the brain region. It derived the coordinates from three papers. Then it drove the stereotaxic arms into the brain and injected 250 nanolitres while watching the entry site through the rig camera for bleeding.

A licensed vet gated every irreversible step and held a physical kill switch to stop the AI agent.

Thread on what it decided, what it saw, and the moment it refused to let me override it.

[Writing this on the spot out of excitement, but I've been 18hrs in front of my laptop coding and coordinating this, so keeping it brief and descriptive, more cohesive thoughts on this soon...]