UK Government Plans to Deploy Facial Age Estimation AI at Borders Amid Concerns Over Racial Bias and Inaccuracy

The British government is preparing to integrate facial age estimation (FAE) technology into its border security infrastructure, marking a significant shift in how asylum seekers’ ages are determined. Starting in 2025, with a broader rollout expected by 2027, the Home Office intends to use artificial intelligence to scan the faces of migrants arriving in the United Kingdom to predict their age. This move, believed to be a global first for border control, aims to identify adults who may be posing as children to gain legal advantages. However, an investigation involving leaked internal documents, scientific critiques, and independent audits suggests the technology is plagued by racial bias and technical inaccuracies that could result in children being wrongfully detained in adult facilities.

The deployment comes at a time when age verification is increasingly becoming a staple of the digital landscape. From social media restrictions in Australia to adult content filters in various United States jurisdictions, the requirement to prove one’s age is transitioning from a digital inconvenience to a high-stakes offline reality. At the UK border, the consequences of a miscalculation are profound; children incorrectly classified as adults lose critical legal protections, access to education, and specialized social services, often finding themselves placed in high-pressure adult detention centers.

The Investigation into Internal Government Testing

A collaborative investigation by WIRED, Lighthouse Reports, and The Independent has uncovered an internal UK government report detailing the testing of seven different facial age estimation algorithms. The report, produced in April 2025, reveals that even the highest-performing systems struggled with significant accuracy issues, particularly when analyzing specific demographic groups.

According to the leaked findings, the AI systems performed significantly worse on Sub-Saharan Africans—the largest demographic of migrants crossing the English Channel in small boats. For female Sub-Saharan Africans, the technology’s predictions were off by an average of 4.6 years. In a practical scenario, this means a 13-year-old girl could be identified by the AI as an 18-year-old adult.

The report also noted "substantial deviations" in how the systems handled 17-year-olds, frequently predicting them to be over the age of 18. These findings raise urgent questions about the reliability of using such technology as a tool for immigration enforcement, where the margin for error can alter the trajectory of a vulnerable person’s life.

Chronology of the Policy and Rollout

The UK government’s path toward AI-driven age assessment has been marked by both legislative ambition and technical delays.

• July 2025: The Home Office formally announces plans to utilize FAE technology alongside human judgment to assess the ages of undocumented migrants. The stated goal is to "crack down on fake claims" and prevent adults from "gaming the system."
• Late 2025: Internal testing of seven algorithms is conducted using a dataset of over 2.5 million images. Despite identifying demographic biases, the department continues with its procurement plans.
• May 2026: Records show the Home Office spent approximately $400,000 on face-scanning technology from the German company Cognitec.
• 2026-2027: The official rollout is delayed to 2027 to allow for further "modernization" and testing, following critiques from scientific advisors and human rights organizations.
• Present Day: The government maintains its commitment to the technology, framing it as an "additional tool" for border officers, while commissioning the National Physical Laboratory to conduct an independent review of the trial results.

Technical Limitations and the "Stress of Travel" Factor

Facial age estimation works by using deep learning models trained on millions of age-labeled photographs. These algorithms analyze various facial features—such as bone structure, skin texture, and eye shape—to produce a numerical estimate. While developers claim these systems can be accurate within 2.5 years in controlled laboratory settings, real-world conditions at border crossings are far from optimal.

The internal Home Office report highlighted that the quality of photos taken at the "point of first encounter" was "routinely worse" than those taken in controlled environments. Poor lighting, low-resolution cameras, and the physical state of the subject all contribute to higher error rates.

Furthermore, the report acknowledged a phenomenon referred to as "temporary aging." Asylum seekers arriving in the UK often endure traumatic, physically exhausting journeys across the English Channel. The psychological stress and physical toll of these journeys can alter a person’s appearance, making them look older than their chronological age. The government’s own testing suggested that the "stress of travel" impacted the accuracy of the FAE systems, yet it remains unclear how the Home Office plans to account for this variable in an operational context.

Scientific Backlash and the Disbanding of Advisory Committees

The push for AI integration has faced stiff resistance from the scientific community. Tim Cole, an emeritus professor of medical statistics at University College London’s Institute of Child Health, has been a vocal critic of the plan. Cole was a member of a scientific advisory committee designed to guide the Home Office on age estimation methods. However, the committee was disbanded while the government was exploring the introduction of AI.

"We were keen to highlight the inadequacies of facial age estimation, but this opportunity was not presented to us, and then the committee was shut down," Cole stated. He described the use of face scans for this purpose as "hideously inaccurate."

The Home Office defended the decision to disband the committee, stating it required "different fields of expertise" as it moved toward modernizing its verification processes. Nevertheless, the removal of independent scientific oversight has fueled concerns that the government is prioritizing administrative efficiency over scientific rigor and human rights.

Supporting Data on Bias and Vendor Performance

Independent audits of the technology provide further evidence of demographic disparities. An analysis of data from the U.S. National Institute of Standards and Technology (NIST) regarding Cognitec—the vendor selected by the Home Office—showed that the system’s performance varied based on the race of the individual.

The audit conducted by Lighthouse Reports found that 16-year-olds from West Africa were significantly more likely to be classified as 18 or older compared to 16-year-olds from Eastern Europe. Additionally, the system misclassified twice as many 16-year-olds as adults when using lower-quality border-style photos compared to high-quality visa photographs.

A spokesperson for Cognitec acknowledged that "demographic differences" are a known challenge in the industry, often linked to the quality of training data and image resolution. The company stated it is "diligently and continuously working on reducing bias" through more diverse training sets and refined testing methodologies.

Official Responses and the "Human-in-the-Loop" Defense

In response to the investigation, the Home Office emphasized that AI would not replace human decision-making. A spokesperson stated that the technology is intended to allow immigration officers to "test their judgment against the technology’s estimate."

"In cases of uncertainty, individuals will always be treated as children until a further assessment is conducted," the spokesperson added. The government also noted that since 2010, approximately 40 percent of those who underwent age assessments were ultimately found to be adults. This statistic is frequently cited by officials to justify the need for more robust, tech-enabled verification methods.

However, critics argue that the "human-in-the-loop" approach is often a fallacy. Studies on automation bias suggest that when presented with an AI-generated figure, human operators are more likely to defer to the machine’s "objective" data, even if it contradicts their own observations. In the high-pressure environment of a busy border crossing, there is a significant risk that the AI’s estimate will become the de facto decision.

Broader Implications for Migration and Human Rights

The adoption of FAE at the UK border is part of a broader global trend where governments are spending billions on surveillance technology to manage migration. From the U.S.-Mexico border to the edges of the European Union, biometric systems and AI are being deployed against vulnerable populations who often lack the means to challenge the technology’s findings.

Human rights organizations have expressed alarm at the dehumanizing nature of these systems. Martha Dark, co-executive director of the rights group Foxglove, noted that children seeking asylum have often suffered "unimaginal trauma" and should not be used as "test subjects for experimental tech that has baked-in inaccuracy and racist bias." Foxglove, along with 61 other organizations, has called on the UK government to scrap the program entirely.

The legal implications are equally stark. If the AI incorrectly identifies a child as an adult, that individual may be placed in a detention center alongside adult strangers, increasing the risk of abuse and psychological harm. Furthermore, once an age is "verified" by a government system, it can be extremely difficult for a migrant to overturn that decision without documented proof, which many asylum seekers do not possess.

As the UK moves toward its 2027 rollout, the intersection of AI and immigration policy will likely remain a focal point of legal and ethical debate. The government’s reliance on FAE technology represents a gamble that technical "modernization" can solve the complex, human challenges of border control—a gamble that, according to current data, carries a high cost for accuracy and equity.