AI surgery is finding value beyond the autonomous robot

Europe’s surgical services are investing in robotics while researchers develop AI for video analysis and limited autonomous tasks. The most useful advances depend on identifying exactly which part of an operation technology changes.

A surgical robot is a highly visible investment. Software that retrieves a difficult moment from an operation, or helps a team interpret its video feed, is less conspicuous. Both can affect surgical practice, but they do different jobs. For hospitals planning their next generation of operating theatres, that distinction has financial and clinical consequences.

Robot-assisted surgery usually describes instruments controlled by a surgeon at a console. AI can add image interpretation, analysis or automated actions, but the presence of robotic arms does not itself establish that artificial intelligence is making surgical decisions. Evidence for an established robotic procedure cannot simply be transferred to a new autonomous function.

The opportunity for European health services is to assemble capabilities around the work that needs improving: access to minimally invasive procedures, support during difficult dissection, useful feedback after surgery and reliable theatre operation. Purchasing a machine is only one part of that task.

Clinical robotics has its own evidence base

The UK iROC randomised trial compared robot-assisted radical cystectomy with intracorporeal urinary diversion against open surgery for bladder cancer. Published in JAMA in 2022, it found median days alive and out of hospital during the first 90 days of 82 in the robotic group and 80 in the open surgery group. The authors described the clinical importance of the difference as uncertain.

This is evidence about a defined surgical approach, comparator and patient population. The robot was surgeon controlled. It is not evidence that autonomous surgery improves recovery, nor does a comparison against open surgery establish the same advantage over conventional laparoscopy.

The distinction matters when a hospital develops its business case. A service converting appropriate open operations to minimally invasive procedures faces a different question from one replacing established laparoscopic work. It should identify the specific patients and procedures likely to change, then examine recovery, complications and resources for that comparison. Otherwise, a benefit demonstrated in one clinical setting can become an unsupported promise across an entire surgical programme.

Conditional adoption creates a practical learning period

In April 2025, NICE allowed five robotic technology options for soft tissue procedures to be used in the NHS during a three-year evidence generation period, subject to conditions. Its assessment excluded prostatectomy, where robot-assisted surgery was already established practice. The recommendation requires evidence on learning curves, resource use and clinical outcomes, with reassessment after the period.

That is a conditional route to use, rather than a finding that every platform is equally valuable in every hospital. NICE highlights substantial introduction costs and the importance of the intended procedures, staffing, training and utilisation. Its assessment also distinguishes the prospects of replacing open surgery from replacing other minimally invasive approaches.

For surgical service managers, the learning period should be planned into the operating schedule. Training lists, proctor availability and staff familiarisation are real uses of theatre capacity. Measuring only the busiest months after a team becomes proficient would obscure the resources needed to reach that point.

An acquisition agreement can reflect this progression. Installation, supervised introduction and established service delivery are different phases, with different support requirements. Those phases should be visible in implementation milestones rather than hidden within a single promised procedure volume.

Registries show use across ordinary surgical practice

The Versius Surgical Registry illustrates another form of evidence. An early analysis reported 2,083 procedures across surgical indications and recorded intraoperative and postoperative outcomes. The study was supported by CMR Surgical and included company-affiliated authors. It documented experience with implementation; it was not a randomised comparison showing superiority over competing systems.

Registry data can help a service examine how results change as teams gain experience and as the case mix broadens. Their value increases when the records contain enough detail to distinguish a more difficult patient population from a technical or organisational problem. Procedure counts alone cannot make that distinction.

For manufacturers, a registry can also provide a structured channel for learning about use after deployment. Hospitals considering participation should establish what information they receive back, how missing follow-up is handled and whether they can analyse their own outcomes independently. Those are practical conditions for making the data useful to clinical teams, rather than merely accumulating a larger total of recorded operations.

AI can make surgical video easier to use

A July 2026 study in npj Digital Surgery developed computer vision models using 683 laparoscopic gallbladder surgery videos from six sources. The system identified surgical steps and retrieved examples of bleeding, bile spillage and thermal injury. Its purpose was to help users navigate relevant parts of recorded procedures for postoperative review. The dataset contained no common bile duct injuries.

This is a different product proposition from automated instrument control. The immediate task is finding material worth reviewing, so a trainer does not have to search an entire recording manually. The study did not establish that deploying the tool reduces complications or shortens the time needed to train a surgeon.

A useful deployment would preserve the surrounding footage, allowing the reviewer to examine what preceded an event and how the team responded. A clip without context can conceal the difficulty of the anatomy or the reason for a decision.

For hospitals, the purchasing question is whether the retrieval system fits an actual review process. If clinicians lack allocated time to discuss the selected footage, improved search may produce an unused library. If review is already part of training, the same capability could support more focused discussion without changing who controls the operation.

The operating room changes the software requirements

A Canadian study published in May 2026 compared cloud and local, or edge, arrangements for delivering surgical AI. Twenty-three end-users tested deployment in a high-fidelity simulated operating room. Cloud users completed more setup tasks successfully, while overall usability scores did not differ significantly. The physical arrangements also produced different ergonomic effects. Cost and carbon estimates were calculated rather than measured as patient outcomes.

The study demonstrates why an AI system’s interface and installation deserve direct evaluation. A model may produce an intelligible overlay, yet its monitor position, cables or startup sequence can make it awkward to use around an operating table. These are characteristics of the deployed system, not just the algorithm.

European buyers can use simulated theatre sessions to compare proposed arrangements before routine adoption. The exercise should include the people who connect and troubleshoot equipment as well as the surgeon viewing the result. It should also test what happens when a feed disappears or the software becomes unavailable. An operational fallback is part of the surgical service being purchased.

Autonomous research tackles a narrower challenge

In July 2025, Johns Hopkins University reported work on SRT-H, a system trained through surgical demonstrations and language-conditioned imitation learning. It performed a lengthy phase of gallbladder removal on pig cadaver material, including a sequence of tasks involving ducts, arteries, clips and cutting. Researchers also tested changes in starting position and visual appearance. This was preclinical work, not an autonomous operation on a living human.

The achievement concerns coordinating actions in a changing visual scene. Its next clinical questions include handling tissue conditions and emergencies beyond those represented in the experiments. A bounded successful demonstration cannot establish how a system would manage every complication within a complete operation.

That makes the handover between automated and human action an important design problem. An automated step needs a clear starting condition, a way to recognise that its assumptions no longer hold and a workable route for the surgeon to regain control. Hospitals evaluating research collaborations should know which of those functions has actually been tested.

Teams determine how the capabilities fit together

The Royal College of Surgeons of England’s December 2025 guidance calls for structured simulator training, supervised cases, competency assessment and formal governance as robotic surgery expands. These are professional recommendations for introducing and sustaining a surgical service.

AI adds further choices to that service. Video review, visual assistance and instrument automation have different users and responsibilities. Each should have a defined purpose in the theatre or training programme, with enough information for staff to understand its limits.

The commercial advantage may therefore come from making one part of surgery work better and connecting it reliably to the rest. Europe does not need to wait for an autonomous operating theatre to pursue that opportunity. It needs procurement and evaluation that can distinguish a useful component from a claim about the whole operation.