UK Plans AI Face Scans To Judge Asylum Seekers’ Ages Despite Known Bias Risks

The UK government plans to start using artificial intelligence to estimate the ages of asylum seekers by scanning their faces, even as its own testing shows that the technology can be inaccurate and biased against key migrant groups.

From 2027, facial age estimation (FAE) systems are set to be introduced as part of age checks at the UK border, according to government plans first set out in 2025. The tools analyse facial features and return an estimated age, which the Home Office says will be used alongside human judgment to help decide whether people claiming to be under 18 should be treated as adults.

The stakes are high as asylum seekers found to be adults can be denied protections given to children and placed in adult-only detention. Many arrive without documents and after dangerous journeys across the English Channel, making age assessments both critical and difficult.

An internal Home Office report from April 2025, obtained by WIRED and Lighthouse Reports in collaboration with The Independent, details testing of seven facial age estimation algorithms on more than 2.5 million images. The report centres on what officials described as the “best performing algorithm” but still documents serious shortcomings.

The system was found to perform significantly worse on people from Sub-Saharan Africa than on other groups. Sub-Saharan Africans have been the largest group of migrants crossing the Channel in small boats in recent years and, according to Home Office figures cited in the investigation, they faced more age assessments in 2025 than other regional cohorts.

For Sub-Saharan African girls, the average age error was 4.6 years. On that basis, the system could assess a 13.5-year-old as an 18-year-old adult. On average, the algorithm also tended to predict that a 17-year-old would be over 18 and it performed worse overall for females, the report says.

The government’s own analysis warned that the technology’s performance could be even poorer in real conditions than in the lab. Most testing was based on relatively high-quality images of documented people. The few photos taken at migrants’ initial encounters at the border were “routinely worse” than follow-up pictures of the same people, and the report could not determine whether the poor image quality or the physical condition of asylum seekers arriving after stressful, physically demanding journeys had more impact on the results.

The report also noted that “temporary aging” related to trauma and travel stress appeared to affect the systems’ accuracy. The US National Institute of Standards and Technology (NIST) has separately found that age-estimation algorithms tend to produce larger errors on lower-quality photos and that accuracy varies with race and other demographic factors.

Disbanded advisory committee and rights concerns

As the Home Office developed its AI plans, it disbanded a scientific committee that had been set up to advise on age estimation methods more broadly. Tim Cole, an emeritus professor of medical statistics at University College London’s Institute of Child Health and a former member of that committee, said members wanted to raise concerns about facial age estimation but were not given the chance before the body was shut down. Cole described the facial scan approach as “hideously inaccurate.”

The Home Office says the committee was dissolved because it needed “different fields of expertise.” In public statements responding to questions about the AI rollout, the department has insisted that FAE will not “replace or overrule human judgment” and will be used as an “additional” tool to support immigration officers. In cases of uncertainty, it says, people will be treated as children until a further assessment is completed.

However, the internal report acknowledged that how the technology would actually be used “in an operational context” was still under exploration. The Home Office has not directly answered detailed questions about how border staff will interact with AI-derived age estimates, what specific training they will receive on the systems’ weaknesses, or what standards will apply to the images used.

Civil society groups argue that the rollout risks embedding flawed automation in a process that already struggles with quality and transparency. Martha Dark, co-executive director of digital rights group Foxglove, said children seeking asylum “have often suffered unimaginable trauma” and “should not be the test subjects for experimental tech that has baked-in inaccuracy and racist bias.” Foxglove and 61 other organisations have signed an open letter urging the government to abandon the plan.

Human Rights Watch researcher Anna Bacciarelli warned that any use of facial age estimation at borders could become “dehumanizing” and normalized over time, describing high levels of risk “in every component of this system.”

Concerns are amplified by past findings about the UK’s existing, human-led age checks. Reports from the UK’s Independent Chief Inspector of Borders and Immigration have highlighted “poor” record-keeping, “perfunctory” visual assessments and a lack of clear explanation of processes to those being assessed. Until 2023, staff conducting age assessments were not given specific training for the task, according to the last inspection report.

It remains unclear exactly which systems the UK will deploy. The leaked Home Office report did not name vendors, though it evaluated seven algorithms. In May 2026, the government spent more than $400,000 on face-scanning technology from German firm Cognitec, one of the companies whose systems had been tested.

An analysis of NIST data by WIRED and Lighthouse Reports found that Cognitec’s age estimation system misclassified twice as many 16-year-olds as being 18 or older when tested on lower-quality border crossing photos compared to higher-quality visa images. The same analysis of NIST scores indicated demographic disparities, including a greater likelihood that 16-year-olds from West Africa would be classed as 18 or older than 16-year-olds from Eastern Europe.

Cognitec declined to comment on its work with the Home Office but said demographic performance differences affect all face-scanning algorithms and are “extremely complex” and often tied to image quality. The company said its algorithms’ bias is “low compared to other algorithms of similar overall accuracy” and that it is working to reduce it by changing testing methods, training approaches and diversifying data.

In controlled lab environments, leading age-estimation systems can predict age within roughly 2.5 years on average. But the Home Office’s own report, alongside NIST’s long-running evaluations, underscore that real-world use is far messier. Performance varies widely by algorithm, gender, demographic group and photo conditions, and lower-quality images such as those captured in busy, poorly lit arrival areas can sharply increase error rates. Some systems have even been fooled using images of video game characters.

Despite these limitations, the UK government has framed the technology as “cutting-edge AI tech” that can help “crack down on fake claims” and stop adults from “gaming the system” by pretending to be children. Each year, tens of thousands of people seek asylum in the UK, and since 2010, about 40 percent of those subject to age assessments have ultimately been classified as adults, according to official statistics.

The Home Office says AI age estimation will allow immigration officers “to test their judgment against the technology’s estimate.” At the same time, it has tasked the UK’s National Physical Laboratory with an “independent review” of testing results and trials. Officials have also floated the idea of using threshold ages such as configuring a system to identify whether someone is under 20 to try to reduce error margins, but have not confirmed whether such thresholds will be adopted or how they would be set.

The UK’s AI-driven age checks are emerging against a broader backdrop of governments worldwide investing heavily in border surveillance technologies, even as rights groups warn that migrants often have little knowledge of how such systems work or how to challenge them. In parallel, facial age estimation has rapidly spread online as a core technology for age-gating social media, adult content and some retail services, including trials in UK bars and shops.

At the border, though, an incorrect guess about someone’s age can reshape the course of their life determining whether they are treated as a child or an adult in one of the most high-stakes environments the technology has yet encountered.