They requested the AI mannequin to decide on a therapy from 19 choices in a spreadsheet. A veterinary surgeon first carried out a craniotomy on the mouse whereas the AI agent noticed. The agent chosen the remedy, mind area, and injection coordinates, then managed the robotic arms and injector needle, in addition to all of the parameters, reminiscent of quantity and velocity.
The staff initially examined Anthropic’s Claude Opus 5.5 mannequin, nevertheless it refused to information the needle into the mouse’s mind, suggesting it would cease halfway by the surgical procedure, which may endanger the animal.
“He did the experimental planning, however when it was time to inject, he mentioned, ‘one thing on my output is telling me I should not do that,’” Santero says, referring to the agent. “That was very fascinating.”
The staff switched to utilizing Claude Sonnet 5, which agreed to proceed. Opus is a heavy-reasoning intelligence specialist greatest for complicated multistep problem-solving, in accordance with Anthropic, whereas Sonnet is healthier suited to on a regular basis duties, although it could be a stretch to name mouse mind surgical procedure a standard chore. (Anthropic didn’t reply to a request for remark.)
Frontier labs have been planning for extra widespread bodily manifestations of AI, and researchers have already begun enlisting brokers to do different issues within the bodily world. In August, Anthropic put out a framework for the way its fashions can work together with the true world. Extra not too long ago, AI engineers put GPT, Claude, and Grok in charge of a car to choose up lunch at In-N-Out Burger. Solely one of many fashions succeeded.
The mind injection process wasn’t with out hiccups both. The researchers had established safeguards, together with requiring the AI to suggest an injection plan for human approval and putting in an emergency button a human may hit that may disconnect the rig’s motors. In addition they restricted the agent’s entry to the underlying gear code to stop it from modifying its personal working guidelines.
