In a hospital outside Paris earlier this year, a surgeon performing a gallbladder removal glanced away from the monitor for a moment. The camera inside the patient’s abdomen didn’t drift. It held steady, tracked the surgeon’s instruments, and quietly reframed the view without anyone touching a joystick. The system making those adjustments was ScoPilot, an AI-powered module built into Moon Surgical’s Maestro robotic platform. In May 2025, the FDA cleared ScoPilot under clearance ID K250984, making it one of the first commercially authorized systems to take partial, real-time control of a task during a live surgical procedure.
That clearance marks a turning point. AI has been reading medical images and flagging anomalies for years, but those tools operate after the fact or alongside a clinician’s workflow. ScoPilot acts during the operation itself, processing surgical video on NVIDIA’s Holoscan computing platform and making continuous micro-adjustments to the laparoscopic camera while the surgeon cuts, cauterizes, and sutures. The distinction matters: this is not a second opinion. It is a machine sharing control of the physical environment inside a patient’s body.
From Camera Control to Full Autonomy: The Research Trail
ScoPilot handles a single, bounded task. But the research pushing toward broader surgical autonomy has been building for years. In a study published in Science Robotics in January 2022, a Johns Hopkins-led team reported that a robot called STAR performed laparoscopic intestinal anastomosis, the reconnection of two severed ends of intestine, under supervised autonomy in pig models. Anastomosis is one of the most technically demanding soft-tissue tasks in abdominal surgery. A single misplaced suture can cause a leak that leads to sepsis.
STAR’s results were striking. The robot matched or exceeded human surgeons on key precision metrics, including suture spacing consistency and tissue leak pressure. The study gave the field its strongest experimental evidence that a machine could handle complex, deformable-tissue work, not just rigid, pre-mapped motions. But the results carry important caveats: porcine intestinal tissue behaves differently from human tissue, the study included no long-term follow-up on complications like strictures or late leaks, and no multi-center human trial of STAR has been published in the four years since.
A separate line of research is tackling the software architecture that would let robots handle longer, more unpredictable procedures. A team described a framework called SRT-H in a May 2025 preprint on arXiv, outlining a hierarchical autonomy system that uses language-conditioned imitation learning to manage extended surgical task sequences. The framework is designed to let a robot recover from mid-task errors and generalize across procedural steps, addressing a core weakness of earlier systems that could only execute rigid, pre-programmed motions. SRT-H has not been tested in clinical settings or peer-reviewed, but it represents a distinct technical approach to the same problem ScoPilot and STAR are solving from different angles: giving machines the capacity to act with partial independence during surgery.
Washington Is Already Drawing the Map
Federal agencies are not waiting for the technology to mature before setting boundaries. The Advanced Research Projects Agency for Health published a request for information on autonomous robotic surgery, laying out a scale of surgical autonomy from Level 0 (no autonomy) to Level 5 (full autonomy without human oversight). The RFI asks pointed questions: What training should surgeons complete before supervising an autonomous system? What supervision protocols should apply at each autonomy level? Where does liability fall when a machine, not a hand, makes the cut that causes harm?
That framing signals Washington views partial surgical autonomy as an imminent regulatory challenge, not a distant hypothetical. But the RFI collects input; it does not establish binding rules. As of June 2026, no federal agency has published specific credentialing requirements for surgeons who oversee autonomous systems, and no formal liability framework exists for AI-guided surgical injuries. Major surgical societies, including the American College of Surgeons and the Society of American Gastrointestinal and Endoscopic Surgeons, have not yet issued detailed practice guidelines for this category of technology.
Researchers writing in Nature Machine Intelligence in 2024 described how general-purpose foundation models could accelerate surgical robots toward greater independence. Yet every currently cleared system still requires a human supervisor with the ability to override the machine at any moment. The gap between what foundation models might enable and what regulators will permit is where much of the tension in this field now lives.
What Hospitals Still Don’t Know
For the administrators and surgeons evaluating whether to adopt these systems, the evidence base remains thin in critical areas. No public data on patient outcomes from ScoPilot-equipped procedures has been released since the FDA clearance. The agency’s 510(k) process confirms that a device meets safety and effectiveness standards for its intended use, but the underlying performance data, including latency measurements, complication rates, and operative time comparisons, has not been made available by Moon Surgical or the FDA. Independent, multi-center trial results do not yet exist for this product.
Cost is another open question. Robotic surgical platforms already carry significant price tags; Intuitive Surgical’s da Vinci systems, the dominant player in robot-assisted surgery, can cost hospitals upward of $1.5 million per unit before instrument and maintenance fees. Moon Surgical has positioned Maestro as a more affordable alternative, but detailed pricing and total cost-of-ownership figures for ScoPilot-equipped systems have not been publicly disclosed. For smaller hospitals and those serving underinsured populations, the economics of adoption could widen existing gaps in surgical access.
The FDA’s broader list of AI- and machine-learning-enabled medical devices continues to grow, with new clearance IDs appearing regularly. But the public index does not always make it easy to distinguish surgical tools from diagnostic aids or administrative software, complicating efforts by hospital leaders to benchmark their own adoption against national trends.
Where the Scalpel Meets the Algorithm
The current record supports a clear, if narrow, conclusion. Limited, task-specific autonomy, such as AI-driven camera control, has crossed the regulatory threshold into commercial use. Fully autonomous execution of complex procedures remains confined to animal labs and research prototypes. Federal agencies are preparing frameworks for a future in which higher levels of autonomy may be feasible, but the detailed rules governing that future have not been written.
What makes this moment different from previous waves of surgical technology hype is the convergence of three things happening at once: a commercially cleared product that acts during live procedures, experimental systems that have demonstrated autonomous soft-tissue surgery in animal models, and a federal government that is actively soliciting input on how to regulate autonomy levels that do not yet exist in clinical practice. Each of those developments, taken alone, would be incremental. Together, they describe a field that has moved from “Can a machine do this?” to “Under what conditions should we let it?”
For patients, the practical takeaway as of mid-2026 is straightforward. If your surgeon uses a system like ScoPilot, the AI is handling a specific, bounded task, camera positioning, under continuous human supervision. No cleared system is making decisions about where to cut, how much tissue to remove, or when to change course. That boundary is real, it is enforced by regulation, and it is unlikely to shift quickly. But the research pipeline behind it is moving faster than most people outside the field realize, and the policy infrastructure meant to govern the next steps is still under construction.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
