The UK Home Office will start using AI facial age estimation on asylum seekers arriving at the British border in 2027, despite an internal government report showing the technology misjudged the ages of female Sub-Saharan Africans by an average of 4.6 years. An investigation produced in partnership with Lighthouse Reports and The Independent obtained the leaked April 2025 document, which evaluated seven facial age estimation algorithms against more than 2.5 million images. The Home Office spent more than $400,000 in May 2026 on face-scanning technology from the German vendor Cognitec, and is proceeding with deployment after delaying an initial July 2025 rollout target.
The stakes are concrete. Asylum seekers who arrive without documents and are wrongly classed as adults lose legal protections afforded to children and can be placed in adult-only detention. A 4.6-year error means a 13.5-year-old girl could be flagged as 18. Since 2010, 40% of people who faced UK age assessments have been classed as adults, according to official statistics, and Sub-Saharan Africans were the largest cohort subject to those assessments in 2025.
The leaked report describes the unnamed "best performing" of the seven algorithms tested. In controlled laboratory conditions, the best systems hit accuracy within 2.5 years. But the same system showed "substantial deviations" on Sub-Saharan African faces and a systematic tendency to predict that 17-year-olds were over 18. Performance was worse on females across the board.
Key facts
- 01UK Home Office will deploy facial age estimation at the border starting in 2027 to assess asylum seekers without documents.
- 02Internal tests on 2.5 million images across 7 algorithms found average errors of 4.6 years for female Sub-Saharan Africans.
- 03The Home Office spent over $400,000 in May 2026 on face-scanning tech from German vendor Cognitec.
- 04Cognitec's system misclassified twice as many 16-year-olds as adults on lower-quality border photos versus visa photos.
- 0561 organizations co-signed an open letter on June 18, 2026 demanding the rollout be scrapped.
Tim Cole, an emeritus professor of medical statistics at University College London's Institute of Child Health, served on a Home Office scientific committee that advised on age estimation methods. The committee was disbanded while the department was exploring the AI rollout. Cole has called the face scans "hideously inaccurate." A Home Office spokesperson said the committee was disbanded because the work required "different fields of expertise."
“We were keen to highlight the inadequacies of facial age estimation, but this opportunity was not presented to us, and then the committee was shut down.”— Tim Cole, Emeritus professor of medical statistics, University College London Institute of Child Health
The internal report's accuracy figures were generated using high-quality reference images. The few photos in the testing data that came from actual first-encounter border conditions were "routinely worse" in quality than follow-up images of the same people, and the report could not separate how much of the degraded performance came from image quality versus the physical condition of asylum seekers after dangerous crossings. Testing by the US National Institute of Standards and Technology has repeatedly shown that lower-quality photos magnify errors in facial age estimation systems.
A WIRED and Lighthouse Reports analysis of public NIST data on Cognitec's system found it misclassified twice as many 16-year-olds as adults when run against lower-quality border photos compared with high-quality visa photos. The same NIST data showed 16-year-olds from West Africa were more likely to be flagged as 18 or older than Eastern European 16-year-olds of the same age. A Cognitec spokesperson declined to comment on the Home Office contract but said "demographic differences in performance apply to all face scanning algorithms" and attributed bias largely to image-quality issues.
Foxglove, a UK rights group, organized an open letter co-signed by 61 organizations and sent to the government on June 18, 2026, asking the Home Office to scrap the plan. Martha Dark, the group's co-executive director, said the technology has "baked-in inaccuracy and racist bias" and that children seeking asylum should not be its test subjects.
The Home Office position is that face scanning will be "additional" to border officer judgment and won't "replace or overrule human judgment." A spokesperson said: "In cases of uncertainty, individuals will always be treated as children until a further assessment is conducted." The department did not answer questions about how officers will weigh an AI estimate against their own assessment in practice, or what threshold of algorithmic confidence will trigger an adult classification.
Facial age estimation has spread quickly in consumer contexts over the past five years — social media platforms, porn sites, UK bars and shops have all piloted it as governments push for online age verification. Researchers have shown some systems can be fooled by video-game character images. Border use is a meaningful escalation: the consequence of an error is not a blocked website but detention.
The UK is the first government known to apply facial age estimation in an asylum context, which puts it in front of an inevitable wave of similar deployments as governments expand surveillance technology against migrants. The product question is whether vendors like Cognitec can close the demographic accuracy gap before the deployment generates a lawsuit large enough to halt it. The market question is whether being the first reference customer for high-stakes border AI is a contract worth winning. On current numbers, the system will misclassify children from the exact demographic most likely to encounter it, and the Home Office has chosen to ship anyway.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
