Software to guess a child's age, bought after its errors on African children were published.
The Home Office awarded a £322,000 contract in May making Cognitec the UK's facial age estimation supplier for asylum age assessments. A June investigation by Lighthouse and Wired, using the Home Office's own figures, found the technology was off by 4.6 years on average for Sub-Saharan African girls and misclassified a majority of West African sixteen-year-olds as adults. Criticism built through late July.
The problem is real and documented. Home Office data obtained by the Helen Bamber Foundation shows 755 children were wrongly identified as adults in 2025 without the tool, and 17 percent of assessments went wrong in the other direction.
Reliably protects children is where it comes apart. The department's own commissioned figures are worst for the demographic group the system will assess most often, so the tool's failures land on the same children the case for it is built around.
Three checks on the error rates across groups, on whether the pre-existing failure figure is independently sourced, and on whether the critics' comparison holds.
1 more couldn't be settled either way, so it's out of both lines rather than counted as a failure.
Being right about what happened and wrong about what it proves are different failures, so these two lines never get averaged into one score. How this works
The logic check
If the claim is right, these things should be true. Here's what we found. Each check says which version of the claim it's testing, the plain one most people would hear or the widest one the same words could carry, and whether it carries the claim's own burden or establishes the background it sits on.
If the tool reliably protects children without a material racial misclassification risk, independent evaluation should show comparable error rates across racial and ethnic groups.
Reading failsA June 2026 Lighthouse/Wired investigation, using Home Office figures, found the mean absolute error for Sub-Saharan African girls was 4.6 years, and Cognitec's algorithm, independently tested by the U.S. NIST Face Analysis Technology Evaluation in April 2026, had higher error rates for people with darker skin and misclassified most West African 16-year-olds as over 18 across four datasets. source
If the Home Office's justification for adopting the tool is accurate, the pre-existing error rate it cites, 755 children wrongly identified as adults without facial age estimation in 2025, should be independently sourced rather than a bare Home Office talking point.
Reading survivesThe 755-children figure is credited to Home Office data obtained by the Helen Bamber Foundation through a freedom-of-information-style disclosure, and separately the Home Office's own figures show 17 percent of assessed migrants were wrongly classified in the other direction, adults assessed as children. source
If critics' comparison of the tool to phrenology is a fair characterization rather than rhetorical excess, the analogy should hold on its own terms.
Can't be checkedBiometric Update's own reporting calls the phrenology comparison, made by a Migration and Tech Monitor co-founder, a dubious characterization, and highlights a different critic's empirical error-rate point as the stronger version of the case against the tool. source
We couldn't settle this either way, and treating it as a failed check would say more than we know.
The receipts
Every quote, checked against its source.
Carrier set
"(d)ecisions informed by a wider range of information, made with greater confidence, will help ensure that vulnerable children get the support and protections they need"
The full scoring
The raw numbers behind the verdict. What R/I/P/E/L and charge stages mean: how we score.
Research verdict: mixed evidence. This is the category the underlying research assigned before we ran the individual checks, and it came back the same on every card in this set.
The Home Office is treating a facial age estimation contract as a confidence-building addition to asylum decisions; four advocacy groups are using the department's own commissioned error data to argue it specifically fails the demographic it is least accurate for. We read the fight as genuinely two-sided rather than settled, since even the reporting outlet pushes back on the critics' strongest rhetorical device, the phrenology comparison, while endorsing their underlying empirical point about error rates.
Source record: hobocode.net Fault Line Report 2026-W31
The Fault Line Report, July 27 to August 2, 2026
All twenty-seven narratives with the full logic test, the mirror check, the R/I/P/E/L factors and the coverage notes on each lane. This page is the plain-language cut; that one is the whole thing, including the polarization methodology that defines every score here.