The Technology at the Border
From 2027, immigration officers at UK border processing centres will have access to a new tool: Facial Age Estimation (FAE), an AI system that analyses a photograph of a person's face and returns an estimated age in seconds. The Home Office awarded a £322,000 three-year contract — to Harlow-based Akhter Computers, with German biometrics specialist Cognitec as the technical subcontractor — beginning June 1, 2026. Live trials are running at Dover's Western Jet Foil processing centre throughout this year; operational rollout is scheduled for mid-2027. The system will be used to assist immigration officers in deciding whether unaccompanied asylum seekers who arrive without documents are children or adults — a determination with immediate legal consequences for housing, detention, education access, and deportation eligibility.
The Strongest Case for Deployment
Before examining what the technology actually does, the government's case deserves a fair reading. False age claims in the asylum system are not a manufactured concern. Under UK law, unaccompanied minors receive local authority care, access to schooling, and legal protections that do not extend to adults. This creates a documented incentive for adult claimants to present as children, which the Home Office says it encounters. Border Security Minister Alex Norris described FAE as a tool to "put a stop to" adults gaming the system and ensure those who misrepresent their age are "identified, detained and removed without delay." The existing Merton Assessment process — which relies on social workers evaluating physical appearance and demeanour — is itself widely criticised as inconsistent, intrusive, and slow. If FAE were accurate and unbiased, deploying it as a supplementary screen would be a proportionate response to a real operational problem.
The problem is that it is not.
What the Home Office's Own Data Shows
The Home Office published detailed guidance on FAE in May 2026. It is candid in ways that should give pause to anyone considering deployment on this population.
Drawing on benchmarking by the US National Institute of Standards and Technology (NIST), the guidance states that even the best-performing FAE systems carry an error margin of approximately 2.5 years at the critical 16-to-18-year-old boundary. Under UK law, a 16-year-old is a child; an 18-year-old is an adult. A ±2.5-year error margin means those two legal categories are not reliably distinguishable by the technology.
The bias profile is worse. The Home Office guidance acknowledges that "FAE performance can vary depending on ethnicity, skin tone, gender, place of birth and quality of input image," and that error rates are "almost always higher for female faces." An investigation by Lighthouse Reports, which reviewed Cognitec's system performance using NIST data and border-quality photographs, found that the contracted tool misclassified more than two-thirds of 16-year-olds as adults. The disparity by origin is stark: fewer than a quarter of Eastern European 16-year-olds were misclassified, compared with more than half of West African 16-year-olds. For Sub-Saharan African girls, the average error reached 4.6 years.
That figure is not marginal. A 4.6-year average error means a 14-year-old girl from West Africa could routinely be estimated as an adult — triggering placement in adult accommodation, potential detention, and accelerated removal from a system that is supposed to protect her. The technology performs worst on exactly the population it will most commonly encounter at UK borders.
The Coalition's Legal Objections
On June 18, 2026, more than 62 organisations — including the EFF, Foxglove, Human Rights Watch, Liberty, Privacy International, Amnesty International, Refugee Action, and dozens of refugee legal support groups — co-signed an open letter to Border Security Minister Alex Norris. They gave the Home Office 21 days to respond and demanded the deployment be halted.
The letter raises three distinct legal challenges. First, the Home Office has not published an Equality Impact Assessment or a Data Protection Impact Assessment (DPIA) for FAE. Both are ordinarily required under the Equality Act 2010 and UK GDPR before deploying high-risk automated processing that affects a protected group. Second, the signatories argue there is no clear lawful basis for any use of asylum-seeking children's photographs in training the models, and no evidence of consent procedures. Third, they note a population-specific accuracy gap: asylum-seeking children often experience trauma-induced physiological changes — malnutrition, dehydration, sleep deprivation, exposure during dangerous sea crossings — that accelerate visible ageing and make their faces systematically atypical of training datasets drawn from other populations.
"To use this for life-changing decisions in refugee processing centres is to introduce an unreliable, untested technology into an already flawed process." — Human Rights Watch, open letter to the Home Office, June 2026
The coalition also flagged a governance concern: the Home Office disbanded its own independent age assessment advisory committee shortly before announcing the FAE plans. Former committee member Professor Tim Cole said he suspected the timing was not coincidental. If an advisory body with relevant expertise was dissolved to avoid scrutiny of a controversial policy shift, that is a serious transparency failure.
The Proportionality Test
Proportionate regulation asks: is the mechanism matched to the risk, is the technology fit for purpose, and are safeguards adequate? FAE fails the second and third tests.
The Home Office insists immigration officers retain full decision-making authority and apply a "benefit of the doubt" approach. That framing is too optimistic. Research on algorithmic anchoring consistently shows that human reviewers are influenced by machine outputs even when instructed not to defer to them. If a system returns "estimated age: 22" for a 16-year-old from Ghana, a procedural instruction to apply benefit of the doubt does not guarantee the officer discounts it.
The real alternative is not to ignore fraud but to invest in human assessment — better-trained social workers, clearer legal standards for Merton-style assessments, and faster resolution timelines. Between July and December 2025 alone, at least 326 children were initially assessed as adults before local authority reviews reversed that finding. That figure demonstrates the existing system fails — but the answer to a flawed human process is not to layer an inaccurate and demographically biased algorithm on top of it.
FAE is not inherently illegitimate. It may become fit for deployment when it can demonstrate equivalent accuracy across demographic groups, when impact assessments are published, and when human oversight is designed to counteract algorithmic anchoring rather than amplify it. The UK Home Office has not met any of those conditions. Proceeding without them does not serve innovation — it shifts systemic risk onto the children least able to challenge it.