The Home Office has published a guide detailing plans to pilot facial age estimation technology throughout 2026, with the aim of providing immigration officers with additional tools for initial age dispute determinations by 2027.
The accessible guide, titled ‘Facial Age Estimation: Using AI to support initial age determinations’, outlines the Government’s plans to introduce Facial Age Estimation (FAE) technology into the UK asylum system.
FAE uses machine learning algorithms to evaluate a person’s facial photo and estimate their age within seconds. The technology, which is being trialled throughout 2026 with the aim of being operationally introduced at UK borders by 2027, provides a rapid and non-intrusive mechanism to assist immigration officers in screening unaccompanied asylum seekers who arrive without identification documents.
Legislative intent and policy change
The integration of artificial intelligence into border workflows builds on commitments made in the May 2025 Immigration White Paper. In July 2025, Dame Angela Eagle, then Minister for Border Security and Asylum, announced FAE as a cost-effective and objective way for immigration officers to test visual judgment in age disputes.
A subsequent Home Office policy paper in November 2025 confirmed that FAE works at a fraction of the cost of physical options. As a result, the government stopped pursuing invasive scientific methods such as bone X-rays and MRI scans. This guide emphasizes that FAE is only a supplementary tool to inform the initial human decision. It is not intended to automate processes, replace human work, or replace comprehensive statutory ‘Merton-compliant’ age assessments.
Sourcing and supplier selection
To facilitate future testing stages and potential rollout, the Home Office published a contract award notice on 29 May 2026, detailing the procurement of the required age prediction algorithms.
The service agreement, worth £322,000, was finalized on 28 May 2026 as a termination of the framework agreement, formally commencing on 1 June 2026 and running until 31 May 2029.
Procurement was split and ordered to two suppliers.
- Actor Computers Limited
- Cognitech Systems GmbH
The contract includes the acquisition of software that can accurately predict the age of a subject, with the primary specified use case being to assist frontline staff in determining the age of individuals encountered without verifiable identity documentation.
Safeguarding and current protocols
Determining chronological age at borders is an important legal and safeguarding obligation. Misidentifying adults as children may allow adults to enter childcare or school settings with minors, while misidentifying children as adults risks placing vulnerable youth in dangerous adult environments.
Under current procedures, immigration officials take a “give the benefit of the doubt” approach to undocumented immigrants who claim to be minors. An individual will only be processed as an adult if two police officers independently determine that the individual’s appearance and demeanor very strongly indicate that the individual is significantly over 18 years of age.
If there is any doubt, they will be treated as children and transferred to local authorities. FAE introduces additional standardized data points into this initial screening stage.
Differences between technical features and facial recognition
This guide addresses notable public misconceptions by contrasting FAE with facial recognition technology. Both systems utilize artificial intelligence, but have completely different operating parameters.
- Facial recognition:
- Built to establish identity, this technology compares live or captured images to a database to determine exactly who a person is.
- Estimation of facial age:
- Built for the sole purpose of estimating age, the system uses algorithms trained on a large dataset of faces with known chronological ages to convert facial features into mathematical values. It doesn’t search for identity data, doesn’t require personal history, and doesn’t connect to or search databases.
Outside of border security, FAEs are increasingly part of everyday life, used by social media platforms to verify user eligibility, online retailers to restrict adult content, and supermarkets to verify purchases at self-checkouts.
Addressing performance and algorithm bias
The Department of the Interior relies on independent performance data from the National Institute of Standards and Technology (NIST), a division of the U.S. Department of Commerce that has evaluated FAE software since 2014.
NIST’s tests use a real-world dataset of approximately 11 million operational images across six global regions, and measure accuracy through metrics such as mean absolute error (MAE) and standard deviation. Since testing began, the top-performing algorithm improved the average MAE from 4.3 years to 3.1 years.
However, this guide acknowledges that there are clear technical limitations. The accuracy of FAE is highly dependent on image quality, age group, gender, and demographic origin. Importantly, at the critical boundary of 16 to 18 years, accurate estimates become less reliable, with major commercial systems having an error of about 2.5 years. Additionally, NIST data shows consistently higher error rates for female faces.
AI systems run the risk of reproducing biases present in the underlying training datasets, so commercial developers regularly train their software on different demographic strata to maximize fairness.
The Home Office conducts rigorous internal testing of industry-leading algorithms across different ethnicities and genders to assess the fairness of the system and establish appropriate margins of error before the technology is deployed on the front line at the border.
The future of border control using technology
The introduction of facial age estimation represents a major shift towards technology-enabled border control, trading in invasive body scans for rapid digital analysis.
While the FAE provides a cost-effective benchmark to strengthen confidence among frontline staff, it cannot serve as a stand-alone solution due to the documented margin of error in the 16-18 standard and vulnerability to demographic bias.
The ultimate fairness and effectiveness of the system will depend on keeping this technology only as an auxiliary aid and ensuring that final welfare decisions remain entirely in the hands of trained human personnel who can apply the legal benefits of doubt.
