A Home Office evaluation of seven facial age estimation systems, obtained by WIRED, Lighthouse Reports, and The Independent, found the technology performs significantly worse on Sub Saharan African migrants, and the government is deploying it at the border next year anyway.
The Home Office tested facial age estimation, an AI that scans a face to predict whether a person is a child or an adult, against seven different algorithms last year. The internal results, obtained by WIRED, Lighthouse Reports, and The Independent, were blunt: the systems regularly misclassify children as adults, and they perform significantly worse on Sub-Saharan African faces than on others. The government plans to begin using the technology at the UK border on asylum seekers next year anyway.
A child the system calls an adult can be stripped of the legal protections afforded to minors in the asylum process and placed in adult-only detention. That consequence sits at the center of a decision that the government's own testing has already flagged, and the rollout has not yet gone live.
The leaked document, a Home Office evaluation of facial age estimation systems, is the basis for the investigation's findings. WIRED and its partners reported that the seven algorithms reviewed in the test each carried serious bias problems, with the steepest accuracy drops on the Sub-Saharan African demographic. That group is not a marginal slice of the affected population. People crossing the English Channel in small boats are predominantly Sub-Saharan African, and that cohort generated the largest share of age assessments raised in 2025.
The system, once deployed, will help decide who counts as a child in the asylum system. The age question matters because unaccompanied minors in the UK asylum process are entitled to a local authority guardian, separate accommodation, and a different legal standard than adults. An adult classification routes a person into the general asylum queue and, in some cases, into adult detention facilities.
Britain's plan, according to the reporting, would make it the first national rollout of facial age estimation against asylum seekers at a border. The Home Office has framed the technology as a way to speed up the slow, contested process of determining whether a young migrant is genuinely a child, an assessment that today relies on social workers and, in disputed cases, on X-ray or MRI bone scans. The replacement has been pitched as faster, cheaper, and less invasive.
The test results undermine each of those justifications on the specific population the system will actually face. If the technology cannot reliably distinguish a child from an adult on Sub-Saharan African faces, then the speed it offers comes at the cost of the most basic safeguard in the process. If the bias gap is structural rather than incidental, the cheaper alternative is cheap only because it shifts the cost onto the migrants who are misclassified.
The decision has not been finalized. Parliamentary committees, the biometrics and surveillance camera commissioner, and judicial review all sit in the path of a live deployment. Continued investigative reporting, including the WIRED-led collaboration that produced the leaked document, also remains a lever. The government's own test results give a clean evidentiary basis for any of those to act before the system goes live.
The next clear watch item is whether the Home Office publishes the test in any form, and whether any of the seven vendors whose algorithms were evaluated have responded to the bias findings. A second, narrower one: the timetable for the border rollout, which the reporting places at "next year," will be the cleanest test of whether the documented evidence changes the policy, or whether it is filed away.