By Himani Jha
On 29 May 2026, the UK Home Office published a guide explaining how it intends to use facial age estimation (FAE) to support initial age decisions of people who arrive in the country claiming to be children. The software converts a photograph of an individual’s face into numbers and returns an estimated age. Testing is scheduled to run through 2026, with implementation at the border planned for 2027. Less than three weeks after the guide was released, Lighthouse Reports, a non-profit investigative journalism organisation based in the Netherlands, published its analysis of a leaked internal Home Office evaluation, finding that the system it examined classified more than two-thirds of 16-year-olds as adults when processing the lower-quality images typical of border settings.
Most of the criticism of the Home Office’s plan to use FAE – which has come from Human Rights Watch, the Refugee and Migrant Children’s Consortium (RMCC) and dozens of other organisations – has focused on concerns over accuracy and bias. Those concerns are well founded. This article argues that another problem lies in where in the asylum process the tool is used. FAE is designed for the first decision a young asylum seeker encounters: the initial age assessment at the border, before the merits of their claim are considered, before a social worker is involved and (usually) before a lawyer is consulted. That decision determines whether the asylum seeker enters the child or the adult track of the asylum system. Deploying an algorithm at this stage of the process risks turning age assessment, which is meant to be cautious and to give the benefit of the doubt, into a gatekeeping mechanism that operates before adjudication. By describing the tool as merely ‘assistive’, it also falls outside most of the rules designed to regulate high-risk artificial intelligence (AI).
The First Decision Sorts Everything After It
Under the Home Office’s current asylum policy, an asylum seeker who claims to be a child is treated as one unless two officers independently determine that their physical appearance and demeanour ‘very strongly suggest’ that they are ‘significantly over 18’. As Free Movement, a leading UK immigration and asylum law resource written largely by practising lawyers, explains, this initial decision requires no social worker, carries no right of appeal and can be challenged only through judicial review. In R (BF (Eritrea)) v Secretary of State for the Home Department [2021] UKSC 38, the UK Supreme Court upheld the policy, holding that the risk of errors in individual cases should be addressed by challenging the decisions themselves rather than the entire policy.
The stakes of that court decision are high. A young person classified as an adult is placed in adult asylum accommodation, is processed under the adult asylum procedure and may face detention. Local authority duties under the Children Act 1989, including the duty to accommodate, are not engaged in the same way (see Irwin Mitchell’s commentary). The young person also loses procedural safeguards associated with child asylum claims, such as an interview with a responsible adult present and a more generous application of the benefit of the doubt when assessing credibility and consideration of child-specific forms of persecution, as set out in UNHCR’s Guidelines on International Protection No. 8: Child Asylum Claims.
The current UK asylum system already makes this error regularly. A report published by the Refugee Council, the Helen Bamber Foundation and the Humans for Rights Network found that more than 1,300 child asylum seekers were wrongly assessed as adults between January 2022 and June 2023. The Independent Chief Inspector’s inspection of age assessments, conducted between July 2024 and February 2025 and published in July 2025, reviewed a sample of 100 initial age decisions from 2023–24. Of the 29 people the Home Office had treated as adults whose cases were later decided by local authorities, 22 were found to be children. Additionally, the RMCC records that at least 326 children were initially misassessed as adults between July and December 2025 alone.
The UN Committee on the Rights of the Child (CRC) has explained why the age decision that officers make on arrival matters so much. In N.B.F. v Spain (2018), the CRC described age determination as being ‘of fundamental importance, as the outcome determines whether that person will be entitled to or excluded from national protection as a child’. The initial age decision is, in effect, a threshold ruling on which protection regime applies. It takes place before the asylum claim is heard, yet shapes how it will be considered.
What the Algorithm Adds to ‘Significantly Over 18’
The Home Office has been clear about why it wants to use FAE. When announcing the plan in July 2025, Minister of State for Border Security and Asylum Dame Angela Eagle said the technology ‘offers a potentially rapid and simple means’ for officers ‘to test their judgements against the estimates produced by the technology’. She added that using FAE could produce results ‘at a fraction of the cost’ of other methods. The tool is presented as a way of checking officers’ initial judgement.
The difficulty is that the judgement being checked is the least defined part of the asylum process. The chief inspector found that ‘what constitutes “significantly” is left largely to the judgement of frontline officers’. An algorithmic estimate fills that discretion with a number. If an officer suspects a young person is over the age of 18, and the software returns the age as 23, the officer’s impression now appears to have scientific support. Scholars of migration technology have described this effect. Petra Molnar, for example, warns of ‘agency laundering’, in which responsibility for a decision is diffused across both human and machine, while Lucia Nalbandian highlights the ‘black box’ problem of algorithmic reasoning that affected individuals cannot see or contest.
The number is also least reliable exactly where it matters most. The Home Office’s own guide acknowledges that ‘even the top systems have an error margin of around 2.5 years’ for those aged 16 to 18, and that performance varies ‘depending on ethnicity, skin tone, gender [and] place of birth’. The internal evaluation reported by Lighthouse found that the system ‘tends to overpredict the age of minors’. Further the Home Office’s April 2025 internal evaluation found that, in testing, more than half of West African 16-year-olds were classified as adults, compared with fewer than a quarter of their Eastern European counterparts. This matters because Sub-Saharan Africans were the largest group undergoing age assessment in the UK in 2025. Raising the age threshold to 24 reduced the overall number of children wrongly classified as adults but widened the disparity: West African minors became ten times more likely to be misclassified than Eastern Europeans.
There are also reasons to expect the system to perform worse on child asylum seekers arriving at the border than it did in testing. In commercial age assurance, where these systems were developed, a so-called challenge age of 25 is used so that anyone who looks younger is asked for documents. Asylum seekers, however, usually cannot produce any documents. A 2025 review in The Lancet Healthy Longevity found that facial-age models tend to predict older ages for people with chronic illness, and that image quality, lighting and camera angle affect performance. For example, young people photographed at the Home Office’s Western Jet Foil processing site in Dover, where initial age decisions are made, will often have just crossed the English Channel after months of travel, poor nutrition and trauma. The RMCC notes that these systems have not been tested on people in these types of conditions.
This raises a question about who bears the cost of uncertainty. The benefit of the doubt places it on the state: if officers are not sure, the person is treated as a child. A tool with a 2.5-year error margin at the decisive threshold adds more uncertainty. Using its output to support a finding of adulthood moves the cost of that uncertainty onto the child, which defeats the purpose of the safeguard. Writing in social work scholarship, Ama-Rose Greaves describes age assessment as operating between care and control. FAE, used this way, moves it towards control.
The Norms the Tool Must Meet
International child rights law is consistent on this point. Under the CRC and the UN Committee on the Protection of the Rights of All Migrant Workers and Members of their Families (CMW)’s Joint General Comment No. 4 (CMW) and No. 23 (CRC) (2017), age assessment should be a holistic, multidisciplinary process, used only when there is real doubt. Any remaining uncertainty should be resolved in the individual’s favour. If there is a possibility that the asylum seeker is a child, they should be treated as one. In its views against Spain, the CRC, beginning with N.B.F., has held that an individual must be treated as a child while their age is being determined, that age-assessment decisions must be open to challenge and that methods with wide margins of error, such as bone X-rays, should not be relied on to override a claim to minority.
The UK has moved in the opposite direction. Section 52 of the country’s Nationality and Borders Act 2022 permits ‘scientific methods’ of age assessment, and the Immigration (Age Assessments) Regulations 2024 specify X-ray and MRI methods, with an individual’s refusal to consent capable of damaging their credibility. These methods have been widely criticised; the Council of Europe has described them as ‘at best an educated guess’, as quoted by the Helen Bamber Foundation. However, those methods at least went through regulations laid before Parliament. FAE is being introduced at an earlier stage, through procurement and published guidance, with no comparable statutory process.
Assistive by Design, Unregulated in Effect
The Home Office’s guide states that FAE ‘doesn’t automate or replace what remains an integral human decision’. That description has legal consequences. The UK has no general AI statute, and its data protection safeguards on automated decision-making attach mainly to decisions based solely on automated processing. Once an officer formally makes the decision, the tool falls largely outside those safeguards, however much the officer relies on its output.
The EU framework shows that this is not just a UK problem. Annex III of the EU AI Act lists AI systems used in migration, asylum and border control as high risk, including systems that assist authorities in examining asylum applications. An age estimation tool used in asylum decisions would plausibly fall within that category. But Article 6(3) of the act lets a provider treat an Annex III system as not high risk where it is ‘intended to perform a preparatory task to an assessment’ – which closely matches how the Home Office describes FAE.
One counterargument is available: Article 6(3) provides that a system ‘shall always be considered to be high-risk’ where it profiles natural persons, and estimating a person’s age from their face is arguably an automated evaluation of a personal characteristic. Even where the act applies, Niovi Vavoula, a scholar of EU migration law and border technologies at the University of Luxembourg, notes that high-risk migration systems are registered in a non-public section of the EU database, while the Digital Omnibus (Regulation (EU) 2026/1744) has postponed the application of the core Annex III obligations until 2 December 2027.
The broader pattern is what migration scholars Tuba Bircan and Emre Eren Korkmaz call ‘function creep’: a technology built to stop teenagers from buying alcohol or accessing adult websites, with a generous error buffer and documentary checks as backup, is being moved into a setting with no such buffer, no reliable access to documents and much higher stakes. By calling it ‘assistive’, the Home Office can make this move without the scrutiny that would apply if it were described as deciding anything.
What Access to Justice Requires
Should the Home Office proceed with this policy, it will be important to build in protective guarantees to mitigate the risk of algorithmic tools acting as a threshold barrier before adjudication. First, FAE should be used strictly in a one-way manner to confirm child status rather than support findings of adulthood, given its inability to reliably identify adults at the threshold age of 18. Second, any adult determination involving algorithmic tools must automatically trigger a comprehensive local authority age assessment, to be conducted by trained social workers to the standards set in R (B) v London Borough of Merton. This should include a holistic inquiry into the young person’s history and circumstances, not appearance alone, with a chance to respond to adverse findings. Interim protection should be provided on the basis that the person may be a child, consistent with the N.B.F. case.
Third, procedural transparency requires that the system’s estimate, margin of error and source photograph be documented and disclosed to the individual, their legal counsel and local authorities to allow an effective challenge. Finally, public accountability requires the publication of performance metrics disaggregated by nationality, sex and age; independent evaluations by the National Physical Laboratory; and statutory impact assessments. Any deployment placed on a statutory footing should also be subject to parliamentary scrutiny.
The debate over whether the Home Office should use FAE at the border is often framed as a choice between a tired officer’s guess and a machine’s estimate. That framing leaves out the legal standard. International law does not ask the state to guess a child’s age more efficiently; it requires the state to treat a young person as a child when it cannot be sure, and to hear them before deciding otherwise. A tool that sorts people before they have been heard, and is least reliable for the children whose ages are most often disputed, does not meet that standard simply because it is being labelled ‘assistive’.
Himani Jha is a Research Associate at National Law University Delhi, focusing on public international law and international criminal law. She holds an LLM in International Law from the West Bengal National University of Juridical Sciences (NUJS), Kolkata.
The views expressed in this article belong to the author/s and do not necessarily reflect those of the Refugee Law Initiative. We welcome comments and contributions to this blog – please comment below and see here for contribution guidelines.