Anthropic Sonnet AI agent performs controversial brain surgery on mouse

Biotech startup Exin Therapeutics has announced the successful execution of an experimental brain surgery on a mouse using an AI agent. The procedure marked a step forward in autonomous in vivo laboratory testing.

Key highlights: - The AI agent analyzed 19 gene therapy vectors and research papers to choose the treatment. - The agent drove robotic arms to inject medication while monitoring via camera. - A licensed veterinarian supervised every irreversible step with a physical kill switch.

Operation details and procedure Co-founder and CEO Gabriel Ocana-Santero shared the procedure on X, explaining how the AI was given a mouse epilepsy model, a spreadsheet of 19 gene therapy vectors, and three research papers. The AI autonomously selected the drug, the brain target, and derived coordinates. It then operated the motorized rig to inject 250 nanolitres of the medication into the brain while watching for bleeding through the rig's camera.

Model selection and safety protocols The team utilized Anthropic's mid-range Sonnet model for the task. Initially, they tried Opus 5.5, which outright refused to participate in a biological experiment, while an older Opus 5 model warned it might stop working mid-operation. To ensure safety, a human surgeon performed the craniotomy, and a licensed vet had to approve every irreversible step alongside having access to an emergency power-cut button.

Ethical questions and implications While the mouse reportedly recovered well, the experiment sparked ethical debates regarding animal testing and accountability when AI makes critical medical decisions. Supporters, however, expressed optimism about applying integrated technologies to find treatments for severe, drug-resistant conditions like long-standing epilepsy.