No Hands on the Scalpel: The Robot That Did the Whole Surgery Itself

A machine trained by watching operations removed a gallbladder from start to finish, no one steering and no one taking over. A year on, surgery is still absorbing what that means, and the team's next target is a living patient.

Picture the operating table. The tools move with calm, deliberate confidence. An instrument reaches in, finds the spot, clips, cuts, adjusts. A voice says, "grab the head of the gallbladder," and the arm obeys, then keeps going on its own, task after task, to the end.

Here is the part that lands in your stomach: no surgeon was driving. No hand was on the controls. The robot decided.

At Johns Hopkins, researchers built a system called SRT-H and pointed it at one of the most common operations on Earth, removing a gallbladder. This is autonomous surgery in its fullest sense: not a machine steadying a surgeon's hand, but one making the calls. The result was published in the journal Science Robotics in July 2025, about a year before you are reading this. That gap is part of the story: a machine crossed a line most of us assumed was years away, and surgery is only now catching up to what it means.

A gallbladder removal, a cholecystectomy, sounds routine, and for a skilled human it is. But "routine" hides a lot of small judgment, and SRT-H handled the whole chain by itself:

That is the landmark: not a steadier hand, but a machine that made the surgical decisions.

You have heard "robotic surgery" for years; hospitals advertise it. So what is the fuss? Today's surgical robots do not decide anything. The famous da Vinci system is a marvel, but it is a puppet: a surgeon sits at a console and moves the controls, and the robot copies their hands, filtering out tremor and scaling big motions into tiny ones. Every choice, where to cut, when to clip, how hard to pull, is still the human's. That is teleoperation, a person at the wheel and the machine mirroring them.

SRT-H is a different animal. It is not mirroring anyone. It looks at the tissue, works out the next move, and does it, and when the anatomy is a little off from what it expected, it adjusts instead of freezing. The distance between "steady hands" and "makes the call" is the distance between a power tool and a colleague.

So how do you teach a machine to make surgical decisions? You do not hand-code every move, because a real body is too variable. Instead the Hopkins team let it watch: videos of surgeons doing the operation, each clip labelled with the step being performed, until the rhythm of the work became learnable.

If that sounds familiar, it should. SRT-H runs on the same kind of AI as chatbots like ChatGPT, a transformer, which is good at predicting what comes next. A chatbot learns it from text; SRT-H learns it from surgery: given this moment, here is the likely next move. Same trick, very different stakes. A chatbot that guesses wrong hands you a clumsy sentence; a robot that guesses wrong is holding a blade against tissue a body cannot spare.

Two things make it feel less like software and more like a trainee:

Now the caveats, because they matter more than the wow:

Then the question no lab result answers: would you let it operate on you? Even a flawless machine needs scaffolding we do not have yet. Who is liable if it makes a mistake, the hospital, the maker, or the engineer who trained it? What must a regulator see first? How does a patient consent to a system that "learned from videos"? The team's next move, from cadaver tissue toward live animals, is where the easy part ends and safety, proof and accountability begin.

Here is why it matters even if you are nowhere near an operating table. The Lancet Commission on Global Surgery estimated in 2015 that about 5 billion people lack access to safe, affordable, timely surgery, often simply because there are not enough trained surgeons where they live.

Now imagine those hands do not have to be human. A machine that performs a common procedure to an expert standard, anywhere you can put one, is one of the few plausible ways to close that gap: a rural clinic, a disaster zone, a ship or a base no top surgeon will ever reach. Expertise that does not have to be flown in, does not burn out, and is not rationed to people near a good hospital.

But the flip side hides in the same sentence: the thing that makes this exciting, a competent surgeon that is not a person, is the thing that unsettles. Nothing changes tomorrow, but a line we had filed under "the future," a machine making surgical decisions alone and getting them right, turns out to be closer than we thought.

In a lab, yes. Johns Hopkins' SRT-H did a complete gallbladder removal, all seventeen steps, with no human at the controls and no takeovers across eight tries, though on pig-cadaver tissue rather than a living patient, and for one procedure only.

No. SRT-H operated only on realistic pig-cadaver gallbladders. The team's stated next step is live animals, where the hard questions of safety, regulation and consent begin.

The da Vinci is teleoperated: a surgeon moves the controls and the robot mirrors their hands, so every decision is the surgeon's. SRT-H makes its own decisions, reading the tissue and adapting when the anatomy differs.

Related from Frontier Signal: this week's deep dive on the AI that runs its own lab, another machine making expert calls.

Frontier Signal explains frontier technology in plain English. This is general information, not medical advice.