June 18, 2026
The UK has turned to AI to determine ages of asylum seekers, and other countries are likely to follow suit. Our investigation finds the technology is deeply unreliable.
Across Europe countries are seeking ways to get tougher on asylum seekers. They also want to save money and speed up processes. AI-powered “facial age estimation” (FAE) tools have developed as a way of preventing children from accessing age-restricted goods and services, such as adult-only online content or cigarettes and alcohol. But the hardening politics around asylum is creating an opportunity for companies developing this technology to pivot to a new market, one with far higher stakes: using AI to determine whether undocumented migrants are under or over 18.
It’s a decision that can have huge consequences: a child wrongly assessed to be over can be put into accommodation for adults, often forced to share bedrooms with adults they don’t know, and they can be liable for detention and forced removal.
Last July, the UK Home Office announced that after several years exploring which scientific and technological methods would best assist the age assessment process, it had concluded that the “most cost-effective” option would be to use AI. Last month, the Home Office signed a contract with Cognitec Systems, a Dresden-based facial recognition company, to supply the age estimation algorithm. It announced that the technology would be used by immigration officers at the border from next year to “crack down on fake claims by small boat arrivals posing as children.”
As the UK begins to trial this technology, and as anti-migrant politics become increasingly mainstream across Europe, other countries will be watching to assess if they, too, can use AI to deter migration. It has already been mooted: an EU Parliament briefing published last July floated the possibility of using AI to assess the age of asylum applicants.
This investigation and audit, based on exclusive leaked material from the UK Home Office and technical assessments from the US National Institute of Standards and Technology (NIST), raises critical questions about whether this type of tech is fit for this new, high-risk purpose.
METHODS
We conducted our own analysis of benchmark data for Cognitec’s system published in April 2026 by NIST, which carries out regular assessments of facial age estimation vendors. Our methodology for analysing the data in the internal Home Office evaluation and the NIST report can be found here.
We were informed by sources of the existence of an internal Home Office evaluation of facial age estimation technology for use on asylum seekers. We requested a copy of the document under Freedom of Information law, but the Home Office refused to disclose it on the basis that it was in the public interest for the department to have a “safe space” to debate the issue “away from external interference and distraction.” We obtained a leaked copy of the report, which tests the technology from seven competing companies.
The UK’s interior ministry disbanded an advisory committee on age assessment days before the AI plans were announced. Two scientists who sat on that committee told us that they had not been consulted about the plans and had grave doubts over the technology. One said he thought the committee had been dissolved in order to prevent them being consulted.
STORYLINES
Our analysis of NIST data revealed that Cognitec’s system misclassified more than two thirds of 16-year-olds as adults when fed photos from the US border, and performed significantly worse on people from African countries. Across all four NIST datasets, the system predicts more than half of 16-year-old West Africans as being over 18, compared with less than a quarter of 16-year-old Eastern Europeans.
The internal Home Office report found that the “best performing” vendor’s FAE technology tended to overestimate the age of teenagers, with 17-year-olds on average predicted to be 18 or over. It also found that it performed worse on female faces.
Crucially, the system described in the report also shows significant bias against Sub-Saharan Africans, with the error rate for this cohort being double that of other groups. This is particularly significant because the top nationalities arriving to the UK on small boats are now Sub-Saharan, and this group also formed the largest group of asylum seekers who had age assessments raised in 2025, according to the Home Office’s own data.
The Home Office report shows that error rates for Sub-Saharan girls are particularly high, at 4.6 years on average, meaning a 14-year-old girl from Sub-Saharan Africa could be predicted to be an adult.
It also notes that many asylum seekers may undergo stress-induced ageing from trauma or travel and that this appears to impact FAE accuracy, although it was unable to systematically test this.
Professor Tim Cole, a former member of the AESAC, the Home Office age assessment committee that was stood down just before ministers announced the use of the age estimation software, said he believed the Home Office dismissed the committee because they didn’t want members to scrutinise the move to AI.
“I suspect [it was] for the reason that we would have been quite damning about it,” said Cole, emeritus professor of medical statistics at UCL’s Institute of Child Health.
The Home Office claimed FAE technology was an “assistive tool” that “does not overrule human judgement”.
Anna Bacciarelli, senior researcher on technology, rights and investigations at Human Rights Watch, said this was of little reassurance given the “well-established phenomenon” of automation bias, whereby people – in this case immigration officers – are prone to trusting a computer’s decision over their own judgment.
She added that the move by the UK set a “dangerous precedent”, saying: “This really is a world-first. If it goes ahead in the UK, other countries at entry points across Europe are likely to follow suit and use of this inaccurate and invasive technology could become widespread.”