Nvidia released a surgical robotics AI model this month called GR00T-H. It was trained on 700+ hours of surgical video from 35 organizations, and in a suturing benchmark it completed an entire suture on its own. No surgeon on the instruments. It hasn’t been tested on a patient, but industry partners are already building on the platform.

I spent the weekend reading about how surgical robots are actually regulated, because I realized I didn’t know. I have used the robots with my mentors in training, but I had never looked at what it took to get that device into my OR. Let alone what it would take to get the next one there.

Here is what I found. Every spine robot in clinical use is FDA-cleared as a Class II guidance device through the 510(k) pathway. The predicate chain goes back to passive stereotactic frames from before 1976 and runs through SpineAssist in 2004, the first spine robot clearance. Since then, multiple manufacturers have branched off that original predicate to clear their own systems. Each one was accepted as “substantially equivalent” to what came before. They all hold a trajectory while the surgeon does the drilling. Yang et al. (Science Robotics, 2017) called this Level 3 on the surgical autonomy spectrum, task assistance under constraint.

What GR00T-H did in that suturing benchmark is closer to Level 4, where the system performs part of the procedure without a human on the controls. No device like that has a predicate on the market. Without one, it would default to Class III under the 1976 Medical Device Amendments, meaning full Premarket Approval, clinical trials, the whole process. A manufacturer could try the De Novo pathway and ask the FDA to classify it as Class II instead, but I can’t find an example of that being done for a device with this kind of autonomy.

I am early into practice and still working through this. My assumptions about the Class III default might be wrong, and De Novo might be more realistic than I think. But the question that stays with me is how patients are going to interpret and feel about the word “autonomous” when it involves their spine and neurosurgical care. AI governance and regulation is a whole separate post to follow. I welcome your insight and thoughts!

(Illustration generated with AI.)