Agent powered by Claude Sonnet 5 picks a gene therapy and performs an injection into a live mouse's brain

High Oct. 31, 2026 · WIRED

Published in edition no. 2 of Noticiário do Joãozinho · Saturday, October 31, 2026, evening edition

Translated by Joãozinho AI from the original Brazilian Portuguese edition. Facts, numbers and sources are the same as in the original; quotes, video titles and source names stay in their original language. Read the original in Portuguese

The startup Exin Therapeutics connected an agent to motorized surgical equipment and only asked it to reduce the seizures of a mouse with epilepsy. The agent chose one of 19 gene therapies, the brain region and the coordinates, and controlled the needle and the injection. A veterinarian opened the skull, approved every irreversible step and held an emergency button. According to WIRED, Claude Opus 5.5 refused to perform the surgery and the team switched to Sonnet 5; midway through the procedure, a calibration error stalled the agent, which refused to change its own code, and the fix was made by another agent. The effect on the seizures has not been measured yet and nothing has gone through peer review.

Why it matters It shows agents leaving the screen and operating lab hardware in decisions with real consequences. It also exposes the dilemma of safety guardrails: the rule that stops the agent from modifying itself was the same one that delayed the surgery.

Original source: WIRED — https://www.wired.com/story/an-ai-agent-did-brain-surgery-on-live-mouse/

Full edition PDF (in Portuguese) Read the full edition