Surgical robots are moving beyond assistance into autonomous suturing, and regulators are scrambling to keep up

Every surgical robot cleared by the FDA today is built around a simple premise: a human surgeon controls it, and the robot translates that control with more precision than a human hand alone can manage. A systematic review of FDA-cleared surgical robots found that 86% operate at the lowest tier of autonomy — pure teleoperation — while only about 6% reach Level 3, conditional autonomy, where the system executes a pre-defined subtask under active human supervision. Full autonomy, where a robot completes a procedure with a surgeon monitoring rather than controlling, has been a research demo, not a clinical category. That line is starting to blur.

What the Johns Hopkins demonstration actually showed

Researchers at Johns Hopkins built SRT-H (Hierarchical Surgical Robot Transformer), a system trained on hours of recorded surgical video rather than hand-coded instructions — the same broad machine-learning approach behind large language models, adapted to robotic control. In testing on a lifelike surgical model, SRT-H completed the 17-step sequence for a gallbladder removal with 100% accuracy across repeated trials, including when researchers deliberately altered the robot's starting position and changed tissue appearance with blood-like dyes to test adaptability. The system also responded to spoken corrections mid-procedure — a surgeon could tell it to adjust, and it incorporated the instruction without a full stop.

That's a meaningfully different capability than existing FDA-cleared systems like Intuitive Surgical's da Vinci platform, which remain teleoperated even in their most automated configurations. SRT-H represents autonomy over an entire multi-step surgical phase, not a single automated maneuver like suture-tying within a surgeon-controlled sequence.

Why this isn't heading to an operating room soon

The gap between a successful lab demonstration and clinical deployment is regulatory, not technical. The FDA's existing draft guidance on robotically-assisted surgical devices explicitly does not cover autonomous robots performing significant parts of a procedure independently of a qualified practitioner — the framework was written for teleoperated systems and hasn't caught up to what SRT-H-class systems can now do. There is currently no defined regulatory pathway for certifying a system that makes intraoperative decisions without moment-to-moment human control.

That absence matters because the questions a regulator has to answer for an autonomous surgical system are fundamentally different from those for a teleoperated one. Liability allocation changes when the robot, not the surgeon, decides how to respond to unexpected anatomy. Training-data provenance becomes a safety question — a system trained on recorded surgical video inherits whatever biases and edge-case gaps existed in that footage, and unlike a human surgeon, it can't necessarily explain why it made a given decision. Post-market surveillance for adaptive, learning-capable systems requires monitoring infrastructure that doesn't exist yet for a device category the FDA hasn't formally created.

The realistic near-term path

Expect autonomy to arrive in surgical robotics the way it arrived in aviation and industrial automation: narrow, verified subtasks first, with human oversight retained for the parts of a procedure where anatomical variation is highest. A system that autonomously handles a well-defined, low-variability segment — closing an incision, navigating a standard anatomical corridor — while a surgeon retains control for judgment-heavy decisions is a far more plausible next FDA submission than end-to-end autonomous surgery. Level 3 conditional autonomy, already present in a small fraction of cleared devices, is the template regulators are more likely to extend incrementally rather than leapfrog.

For hospital systems and device makers watching this space, the practical takeaway is that the bottleneck has shifted. The machine-learning capability to execute a full surgical sequence reliably now exists in a research setting. What doesn't exist is a regulatory framework that can evaluate it, and building that framework — defining acceptable failure rates, human-oversight requirements, and liability structures for an autonomous surgical device category — will likely take longer than it took to build the robot.