From 2027, the UK Home Office plans to introduce facial age estimation technology at the border as an additional tool for assessing the age of individuals whose claimed age is disputed. Although the final decision will formally remain with immigration officers, the use of algorithmic age estimates raises a critical question: what happens when technological uncertainty bears directly on whether a person is treated as a child or an adult?

This blog argues that facial age estimation should not displace the benefit of the doubt owed to individuals whose age remains uncertain. Where an algorithmic estimate cannot reliably distinguish between childhood and adulthood, particularly around the critical sixteen-to-eighteen boundary, uncertainty should operate in favour of child protection rather than against it. Otherwise, a technology presented as merely supporting human decision-making may, in practice, become highly influential in determining access to the safeguards attached to childhood.

The problem is not simply that facial age estimation can be wrong. It is that accuracy is lowest precisely where the legal distinction matters most. A recent UK Parliament briefing reported that leading systems show a mean absolute error of around 2.5 years at the critical sixteen-to-eighteen boundary. That figure does not predict the error in every individual case. It does, however, make the misclassification of sixteen- and seventeen-year-olds a foreseeable risk.

The consequences are substantial. Provisional Home Office statistics published in July 2026 show that, among 1,885 people initially assessed as adults between July and December 2025, 326 were later found to be children, while another 377 cases remained pending. These figures concern the existing age-assessment process, before facial age estimation becomes operational. They already reveal the fragility of initial decisions. A wrong classification can remove a child from the safeguards attached to childhood. The person may be placed in adult asylum accommodation, processed under adult procedures and separated from child-welfare services at the very moment protection is most needed. A later correction cannot fully undo the fear, exposure or loss of support caused by the first decision.

This is not merely an administrative problem. Under Article 3 of the UN Convention on the Rights of the Child, the child’s best interests must be a primary consideration in administrative decisions affecting them. Section 55 of the Borders, Citizenship and Immigration Act 2009 likewise requires immigration functions to be exercised with regard to safeguarding and promoting children’s welfare.

The UN Committee on the Rights of the Child has also made the relevant principle clear: where uncertainty remains after an age assessment, the individual must receive the benefit of the doubt and be treated as a child. The Home Office’s own guidance reflects this approach, stating that where the threshold for treating someone as clearly adult is not met and doubt remains, the person should be treated as a child pending further assessment. Facial age estimation must therefore operate, if at all, within this protective presumption. It should not become evidence capable of tipping an uncertain case into adulthood.

The promise of “human control” is not enough. Numerical outputs can appear fast, scientific and objective, encouraging officials to give them disproportionate weight. The real question is not simply who signs the decision, but whether the officer understands the system’s limitations, can depart from its estimate and must explain what weight was given to it.

Before deployment, the government should publish independent validation results, disclose the uncertainty attached to each estimate and create an accessible route to challenge its use. Most importantly, an algorithmic estimate should never displace the benefit of the doubt owed to a possible child.

At the border, an error of two years is not merely technical. It can determine whether a vulnerable person is protected as a child or exposed as an adult.