The NHS extends its autonomous skin cancer screening AI to 9 million people, already validated on 250,000 real patients

By Business Wire · October 7, 2026.

Every hospital in the NHS's South East England region —18 trusts, with potential coverage of 9 million people— will from this autumn adopt DERM, the artificial intelligence system from British firm Skin Analytics, to carry out the initial triage of suspected skin cancer. According to the company itself and NHS England, it is the largest deployment of autonomous medical artificial intelligence in the history of the British public health system.

DERM is not a chatbot or a language model: it is a computer vision system trained specifically to identify patterns in images of skin lesions. It is, according to Skin Analytics, the only device of its kind certified for autonomous use under Class III CE marking, the most demanding level of regulatory scrutiny for a medical product. It has been integrated into NHS dermatology pathways for more than six years and already operates in 25 English trusts; the announcement extends its use to the 18 trusts of the South East region.

The company cites figures from six years of real-world use: it says it has assessed more than 250,000 patients and found more than 25,000 cancers, confirmed by histopathology (the microscopic analysis of tissue that is the gold standard in oncology). According to its post-market surveillance reports, DERM correctly identifies around 98% of genuine skin cancers —sensitivity— and correctly rules out cancer in 99.5% of benign cases —specificity—. These are two different metrics and it is worth not confusing them: one measures how many real cancers it detects, the other how many healthy patients it can discharge without a false alarm. An independent study cited by the company concludes that its performance matches or exceeds an in-person assessment by a dermatologist.

The practical aim is to ease a waiting list that keeps growing. According to NHS England South East's projections for the three years of the rollout, up to a third of urgent referrals for suspected skin cancer could be resolved —with the patient safely discharged— before the first face-to-face appointment with a specialist. That would free up, by its calculations, some 13,500 consultant hours a year, the equivalent of 27,000 additional appointments in the region. Every patient with a benign lesion that DERM rules out at triage is an appointment a dermatologist does not have to fill, one left free for someone who does have a real cancer or conditions such as eczema or psoriasis, currently competing for the same pool of appointments.

'The only health authority in the world that has taken autonomous AI into cancer pathways at this scale is the NHS, and that is a real milestone that has happened here,' said Neil Daly, founder and chief executive of Skin Analytics. Daghni Rajasingam, regional medical director for NHS England in the South East, framed the move within the British Government's ten-year health plan, noting that using AI alongside clinical judgement gives 'peace of mind' sooner to those who do not have cancer and prioritizes those who do need it. The rollout comes after Northern Ireland recently approved the use of DERM for its population of 1.9 million.

Beyond the press release, what really matters here is traceability: few AI deployments in healthcare arrive with six years of real-world use, published sensitivity and specificity figures, and explicit regulatory certification to act without human supervision on the decision to discharge. In a sector saturated with pilots announced with great fanfare, DERM is already clinical routine in 25 English trusts. Those 250,000 patients and 25,000 histopathology-confirmed cancers are exactly the kind of evidence that separates demonstrated capability from marketing, and they should be the yardstick for the next 'autonomous medical AI' announcement that comes along.

Even so, it is worth keeping sight of what this technology does not solve. The very argument for adopting it —'there are not enough dermatologists to manage the referrals,' in the words of Dr. Alexandra Kemp, of Buckinghamshire Hospitals NHS Trust— describes a structural staffing problem that AI alleviates but does not cure. Automating triage frees up specialist hours, but it does not train a single new dermatologist; if demand keeps growing faster than human capacity, the relief will be temporary. And delegating to an algorithm the phrase 'you don't have cancer, you can go home' shifts the question of clinical responsibility: when that margin of error materializes in a real person, the system will need clear protocols on who reviews the case and how the patient is brought back into the pathway if symptoms persist. The press release does not spell out that mechanism, and it is the question any deployment on this scale should answer with the same transparency with which it publishes its performance figures.

In the long run, however, this is the kind of advance that underpins our underlying thesis on AI in medicine: early detection is by far the factor that most determines survival from skin cancer, and a system able to screen 9 million more people without waiting for the thousandth dermatologist to be trained is a concrete —not speculative— way of moving towards a health service that catches disease before it becomes serious. No artificial general intelligence is needed to achieve it: a narrow model, well regulated and audited over six years, is enough. If the pattern repeats —long clinical validation, demanding certification, gradual rollout—, it is reasonable to expect that what is today mole triage will become, over the next decade, routine screening for several types of cancer at the primary care level.