In 2024, surgeons and care teams used Intuitive Surgical's da Vinci platforms in nearly 2.7 million procedures worldwide, an increase of 17 percent over 2023, bringing the cumulative total since the system's introduction to roughly 17 million procedures across more than 10,600 installed systems. That is one of the largest deployments of robotics in any operating room anywhere — and it is worth pausing on what, mechanically, is happening in every single one of those procedures. A human surgeon sits at a console, moves hand controls, and the robot scales, filters tremor from, and relays that motion to instruments inside the patient. The robot does not decide where to cut, where to place a suture, or how to respond to bleeding. It is a very sophisticated remote control.
That distinction matters because the public conversation about "AI in surgery" tends to conflate two very different things: robots that are teleoperated by a surgeon in real time, and robots that make independent surgical decisions. The commercial reality in 2026 is that almost every FDA-cleared surgical robot on the market — da Vinci, Ion, Mako, ROSA, and others — sits at the teleoperated end of that spectrum. A 2024 systematic review of FDA-cleared surgical robots in npj Digital Medicine found that the overwhelming majority of cleared systems offer no autonomous decision-making at all; autonomy, where it exists commercially, is generally limited to constrained sub-tasks like following a pre-planned bone-cutting path, not to independent surgical judgment.
The six levels, and where real systems sit
The framework most researchers use to talk about this comes from Yang and colleagues' 2017 paper in Science Robotics, which laid out six levels of surgical robot autonomy, borrowing structure from the SAE levels used for self-driving cars. In their scheme:
- Level 0 — No autonomy. The robot has no decision-making role; it is a pure teleoperation or manual-control tool. Da Vinci sits here.
- Level 1 — Robot assistance. The robot provides physical assistance during a task the surgeon still directs — for example, holding a plane steady or damping unwanted motion.
- Level 2 — Task autonomy. The robot autonomously executes a specific, narrow sub-task under continuous surgeon supervision (some bone-milling and stapling systems approach this).
- Level 3 — Conditional autonomy. The robot can plan a task and adapt that plan during execution, with the surgeon approving before and monitoring throughout.
- Level 4 — High autonomy. The robot plans and executes a sequence of surgical tasks with minimal human input.
- Level 5 — Full autonomy. No human needs to be in the loop; the robot performs an entire operation independently.
Every commercially deployed system that operates on humans today lives at Level 0 or Level 1. Nothing on the market plans an operation or adapts a surgical strategy without a surgeon driving each step.
What STAR actually demonstrated — and what it did not
The clearest research-stage counter-example is the Smart Tissue Autonomous Robot, or STAR, developed by a Johns Hopkins team led by Axel Krieger. In a study published in Science Robotics, STAR performed laparoscopic intestinal anastomosis — reconnecting sections of soft tissue — in living pigs with minimal human guidance, using its own imaging and suture-planning to place stitches after a surgeon exposed the tissue and approved the plan. A follow-up study reported the robot completing a more complex, realistic version of the procedure with even less human intervention. These are genuine, peer-reviewed advances, and they sit meaningfully higher on the autonomy scale than any cleared clinical device. But it is important to be precise about scope: STAR has been demonstrated on soft tissue, in animal models, in controlled laboratory conditions, with a human still exposing tissue and approving each plan before execution. It has not been cleared for human use, has not operated without a supervising surgeon in the loop, and has not been tested on the anatomical and physiological variability of real human patients across a hospital population.
Why Level 4 and 5 are further away than the demos suggest
The gap between a research demonstration and a Level 4 or 5 clinical product is not primarily an engineering gap — it is a liability and regulatory gap. If an autonomous robot makes a surgical decision that leads to harm, current frameworks do not cleanly answer who is responsible: the hospital that deployed it, the surgeon who was nominally supervising, the manufacturer that built the perception and planning software, or the institution that validated it. Legal and bioethics scholars writing in Science Robotics have flagged this as an open problem that predates and will likely outlast the current wave of AI hype in surgery — existing product-liability and medical-malpractice law was built around a human decision-maker, and does not map cleanly onto a system that plans its own actions. Regulators have not published a clearance pathway for a Level 4 or 5 device, and no manufacturer has filed for one.
What actually changes in the next few years
Intuitive's own FY2025 results, reported in January 2026, extend the trend rather than change its shape: procedure volume climbed to more than 3.2 million cases (up 19% year over year), with the da Vinci installed base surpassing 12,100 systems worldwide. None of that growth changes where these systems sit on the autonomy scale — every one of those 3.2 million procedures was still teleoperated, a human at the console the whole time. On the research side, the Johns Hopkins SRT-H system's 2025 demonstration remains the most advanced public benchmark: still animal-tissue and simulation work, not human trials, with the team explicit that clinical translation would require its own regulatory pathway. No FDA autonomy-tier classification or liability rule has since been finalized to govern what happens if a more autonomous system moves toward the clinic.
The realistic near-term trajectory is not a jump to autonomous surgery — it is a slow, task-by-task migration up from Level 1 toward Level 2 in narrow, low-variance sub-tasks: autonomous camera positioning, autonomous suturing on standardized tissue segments, AI-assisted anatomy identification that a surgeon still approves in real time. Each of those steps will be validated, cleared, and adopted independently, with the surgeon remaining the accountable decision-maker for the operation as a whole. The more consequential AI story in the OR right now may not be autonomy at all, but perception and judgment support — systems that flag anatomy, warn of complications, or improve a human surgeon's consistency — because that is where the evidence, the regulatory pathway, and the liability model already exist. Genuine autonomy will arrive, if it does, procedure by procedure and years after the demonstrations that make headlines today.