No, an AI did not perform brain surgery in London

Here's the thing the headline skips past: the AI in this story never touched a scalpel, never made a decision, and as far as the Guardian's writeup tells it, was never even given a name. What it did was color in a video feed. That's a real and useful thing. It is also a much smaller claim than "AI-assisted operation to remove brain tumour" wants you to think.

The actual case is genuinely good news. Rhys Hibbert, 48, from Bedfordshire, had an 11mm tumor on his pituitary gland, diagnosed in 2024, that had gotten bad enough by this year to threaten his sight. In May, surgeons at the National Hospital for Neurology and Neurosurgery in London operated on him, and a system trained on "hundreds of surgical videos" watched the live camera feed during the procedure and color-coded nearby anatomy, so the surgeons could see the nerves and blood vessels they needed to avoid without eyeballing it themselves. He walked without a stick within a week and went back to work as a customer service manager. Good outcome, real trial, funded by the NIHR. None of that is in question.

What is worth pausing on is what the system actually did. Dr. Sophia Bano, the UCL associate professor who led the technical build, describes it as something that helps "recognise critical anatomy, surgical instruments and tissue interactions in real time." That's a segmentation and classification model. It looks at pixels and labels them: nerve, vessel, gland, instrument. It doesn't plan the incision. It doesn't decide anything. The surgical team, per the hospital's own framing, "remained in full control" the entire time. If you wanted an analogy, it's closer to the offside line a broadcast overlays on a football pitch than to anything resembling autonomous surgery. The line doesn't play the game. It just makes something hard to see easier to see.

I don't think that's a knock on the achievement. Pointing a camera into a space where the pituitary gland, the optic nerves, and a handful of blood vessels are all packed within millimeters of each other, and having software reliably tell a nerve from a blood vessel from a tumor margin in real time, at video frame rate, without lag that would be dangerous mid-procedure — that's a hard computer vision problem. Health officials aren't exaggerating when they say a millimeter the wrong way can cause blindness, stroke, or worse. Getting that classification right, live, in an actual operating theater rather than a demo reel, is the kind of unglamorous engineering that doesn't sound impressive in a headline and is exactly why it should be trusted more than the headline.

Here's my honest pushback on my own framing, though. One patient is one patient. This was structured as an NIHR-funded clinical trial, which is the correct and conservative way to introduce something like this, and I'd rather see a hospital run it that way than rush to press with a bigger claim. But it also means we have an n of one, no published accuracy numbers, no false-negative rate on what the system missed or mislabeled, and no comparison against a surgeon operating without the overlay on a matched case. Prof Mike Lewis at NIHR calls it "pioneering surgery," which is fair for a first-in-patient use. It is not evidence the tool generalizes yet.

The reader takeaway, if you're the kind of person who reads stories like this because a family member is facing brain surgery: ask your surgical team directly whether they use intraoperative imaging or navigation assistance, and if so, whether it's been through a named clinical trial with published results, not just "AI." The tool matters less than the evidence behind it, and right now the evidence behind this specific tool is one patient.

Further reading

• The Guardian — London neurosurgeons perform first successful AI-assisted operation to remove brain tumour — primary source, all quotes and details above are sourced from here