UCL's Real-Time Surgical AI Left the Lab for Its First Live Brain Operation

A brain tumour was removed at the National Hospital for Neurology and Neurosurgery in London in May 2026 with an AI system reading the live surgical feed. The details were held back until August 27, 2026, while the patient recovered.

The patient, Rhys Hibbert, had an 11-millimetre tumour on his pituitary gland. Left untreated, it would have continued compressing his optic nerves toward blindness.

The operation

Surgeons at NHNN, part of University College London Hospitals, performed the procedure. It was part of a clinical trial funded by the National Institute for Health and Care Research and Google.

The Royal College of Surgeons, the EPSRC, and Wellcome also backed the trial.

Hani Marcus, consultant neurosurgeon at NHNN, performed the surgery. Surgical resident Danyal Khan led the AI work on the case.

The AI had already been used in the research setting and as a surgical training tool. This was its first live use on a patient.

What the system actually did

The distinction that matters here is the input.

Most surgical navigation tools work from pre-operative scans captured before the first incision. This one analysed the live endoscopic video feed in real time as the anatomy was encountered.

The pituitary lies within a millimetre of the carotid arteries and the nerves that control vision.

The system was trained on a large library of annotated endoscopic pituitary surgery videos, learning to recognize those structures along with instruments and instrument-tissue interactions.

It runs on an NVIDIA Clara IGX platform built for real-time AI in medical device settings.

It did not cut, suture, or direct any action. The surgical team stayed in full control throughout.

That places it firmly on the informational end of the surgical AI spectrum, closer to an expert second set of eyes than to anything resembling autonomy.

Sophia Bano, the UCL Hawkes Institute researcher who served as technical lead on the system, said the model's exposure to hundreds of prior surgical videos gives it a breadth of pattern recognition no individual surgeon accumulates early in a career.

The outcome

Hibbert's vision improved immediately after waking. "When I came round… I could see everything in the room clearly," he said.

He was walking independently, without glasses or a stick, within a week.

Marcus credited the trial participants and wider team for what he called an important first step globally in using homegrown UK AI to reduce surgical risk.

Why it's worth tracking

This was not a robotic system, and it made no decisions. It was a perception layer, reading tissue the surgeon was already looking at and flagging what mattered before human eyes might catch it.

That is a narrow, deliberately bounded role.

It is also the kind of claim that is straightforward to verify against a surgical outcome, which is precisely what makes it a credible first step rather than a marketing one.

Source: UCLH