An investigation by WIRED and Lighthouse Reports, conducted with The Independent, obtained a leaked Home Office report detailing tests of seven facial age estimation systems. According to the document, the highest-performing algorithm still produced notable errors, particularly when evaluating people from Sub-Saharan Africa.
The report found that the system’s estimates for female Sub-Saharan Africans were off by an average of 4.6 years. Based on that margin, a girl aged 13.5 could potentially be classified as an adult. The same report also indicated that the algorithm generally tended to estimate 17-year-olds as being over 18 and delivered weaker performance for female subjects.
The findings are especially significant because migrants from Sub-Saharan Africa represent the largest group crossing the English Channel in small boats in recent years and accounted for more age assessments in 2025 than people from other regions, according to Home Office data.
Critics argue the results highlight risks associated with deploying AI in immigration decisions. Tim Cole, an emeritus professor of medical statistics at University College London’s Institute of Child Health and a former member of a government advisory committee on age estimation, said concerns about the technology were not adequately addressed before the committee was dissolved.
“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 said. He also described the face scans as “hideously inaccurate.”
The Home Office said it is modernizing age verification processes through testing of facial age estimation systems while maintaining safeguards. “We have rigorous processes in place to verify an individual’s age and are working to modernize these through the testing of fast and effective facial age estimation technology,” a spokesperson said. The department added that the advisory committee was disbanded because it required “different fields of expertise.”
Facial age estimation systems have become increasingly common in online age verification programs used by social media platforms, pornography websites and retailers. The technology analyzes facial characteristics and uses AI models trained on large datasets of age-labeled images to estimate a person’s age.
While leading systems can achieve relatively accurate results under controlled conditions, performance can vary based on image quality, gender and demographic characteristics. The Home Office’s testing relied largely on high-quality images of documented individuals, and the report acknowledged that real-world performance could be lower.
The document also highlighted challenges involving photographs taken during initial encounters with asylum seekers. According to the report, those images were often of substantially lower quality than photos collected later in the process. Researchers were unable to determine whether poor image quality or the physical condition of migrants had a greater impact on age estimation results.
The report further noted that factors including trauma and the stress of travel appeared to affect the technology’s performance. It concluded that additional research was needed to understand those effects.
Advocacy groups have criticized the proposed use of the technology in asylum procedures. “Children seeking asylum have often suffered unimaginable trauma,” said Martha Dark, co-executive director of Foxglove. “They should not be the test subjects for experimental tech that has baked-in inaccuracy and racist bias.”
Foxglove and 61 other organizations recently sent an open letter urging the government to abandon plans to introduce facial age estimation in asylum assessments.
Questions have also emerged around the technology ultimately selected by the government. The Home Office spent more than $400,000 on face-scanning technology from German company Cognitec. Although Cognitec was among the vendors whose systems were tested, the leaked report does not identify which algorithm produced the results described in the document.
Separate analysis of public testing data cited by WIRED and Lighthouse Reports found that Cognitec’s system classified significantly more 16-year-olds as adults when evaluated using lower-quality border-crossing images than when tested on higher-quality visa photographs. The analysis also identified demographic differences in performance, including a greater likelihood that 16-year-olds from West Africa would be classified as adults compared with 16-year-olds from Eastern Europe.
Cognitec declined to discuss its work with the Home Office but acknowledged that demographic performance differences affect facial age estimation systems broadly.
“The reasons for bias are extremely complex and often related to image quality issues,” a company spokesperson said.
“The bias of Cognitec algorithms is low compared to other algorithms of similar overall accuracy, and be assured that we are diligently and continuously working on reducing bias by developing specific testing methodologies, designing loss functions in our network training, and by diversifying the training and testing data,” the spokesperson added.
The Home Office has maintained that facial age estimation will serve as an additional tool rather than a replacement for human judgment. The department said individuals will continue to be treated as children whenever uncertainty exists until further assessments are completed.
Officials have also commissioned the UK’s National Physical Laboratory to conduct an independent review of testing and trial results. The government has publicly discussed the possibility of using age thresholds to reduce classification errors, though it has not confirmed whether that approach will be adopted.
Human rights advocates remain concerned that introducing the technology at borders could normalize AI-assisted decision-making in situations where errors carry serious consequences.
“Over time there’s a real risk that this will become entrenched,” said Anna Bacciarelli, a senior researcher at Human Rights Watch. “There’s so much risk in every component of this system that it’s really just not worth pursuing to be able to say that you’re using AI to tackle migration.”
This analysis is based on reporting from Wired.
Image courtesy of TripMyDream.
This article was generated with AI assistance and reviewed for accuracy and quality.
