AI triaged suspected skin cancer alone, and rarely missed

Clinical takeaway: Autonomous AI triage moved into routine cancer-pathway service in this UK deployment, with missed cancers rare and none linked to harm so far. Watch for similar efforts to start to emerge in the US. 

Suspected skin cancer referrals keep climbing while most turn out benign, so dermatologists spend scarce clinic time ruling out disease while patients with cancers are forced to wait their turn. Wait times for a dermatology appointment can stretch months in parts of the US, and the specialty's workforce isn't growing fast enough to close the gap. 

Tele-dermatology has stretched specialist reach, but not far enough to meet demand. AI tools have matched dermatologists on diagnostic accuracy in study after study, always with a clinician making the final call. Whether the technology can safely make that call itself, at scale, remained untested. A new report from two UK hospitals delivers what its authors call the first large-scale, prospective look at fully autonomous AI working inside a live referral stream. 

AI discharged 31% and 25% of patients at the two hospitals with no clinician review, and tele-dermatologists released roughly another quarter at each site. That left only a fraction of referrals needing a face-to-face appointment. Consent for autonomous decision-making ran high: 86% of patients agreed to let the system act alone. 

The safety record rests on six missed cancers among 8,391 patients: five basal cell carcinomas and one melanoma in situ, caught through post-market surveillance with no adverse outcomes identified in available follow-up. In a national dataset that includes both sites, sensitivity for malignancy topped 98% for invasive melanoma, squamous cell carcinoma, and basal cell carcinoma, with specificity of 72.1%. 

The autonomous pathway cut routine follow-up to 12% of patients, versus 27% with standard teledermatology, and biopsy rates ran 26% to 29% versus 43% with conventional face-to-face care. The researchers put clinician time saved at 2,851 hours over 16 months, a roughly 62% capacity gain they translate to more than 8,500 additional appointments. 

The device, a CE-marked Class III autonomous AI, went live at the two hospitals in December 2024 and July 2025 after pilot validation with teledermatologist second reads. Referred patients had standardized clinical and dermoscopic images of each lesion captured by smartphone. From there the AI classified each lesion, discharging benign cases with self-monitoring guidance; higher-risk cases, along with all Fitzpatrick type V/VI skin, went to a dermatologist for remote review. Outcomes were tracked prospectively with histopathology as the reference standard for malignant lesions.

The six missed cancers surfaced through post-market monitoring, and the researchers frame that continued vigilance as a condition of autonomous deployment. Before wider adoption the authors call for replication in larger populations and in health systems built differently from the NHS, because imaging infrastructure, referral habits, and liability rules all differ meaningfully. 

"We believe the greatest value of autonomous AI lies not in the technology itself, but in the specialist capacity it unlocks," said lead author Lucy Thomas, consultant dermatologist at Chelsea & Westminster Hospital NHS Foundation Trust. "Every hour saved reviewing low-risk lesions can be reinvested in patients with skin cancer, helping them access timely treatment to improve prognosis, and in patients with severe inflammatory skin disease, where earlier access to specialist care and effective treatments can transform quality of life." 

Source: Thomas L, et al. (2026 Sep 30) EADV Congress 2026, Abstract AS-2145. Autonomous AI triage in urgent skin cancer pathways: Real-world safety, diagnostic performance and system impact in 8,391 patients