AI tool will lead to more child refugees being treated as adults, charity warns
Rights groups and children's charities are warning that new AI-powered facial age-estimation technology, which the UK government plans to introduce to screen migrants, will lead to more child refugees being wrongly classified as adults. Critics say the underlying models show racial bias, disproportionately overestimating the ages of Black children from conflict zones, meaning unaccompanied minors could be housed with adults and exposed to detention, removal and the risk of violence linked to protests at asylum accommodation.
The Home Office has contracted German firm Cognitec to supply the technology, despite an investigation by Lighthouse Reports finding the system biased against sub-Saharan Africans, and government acknowledgement that even the best systems carry a 30-month margin of error. Charities including the Humans for Rights Network and Helen Bamber Foundation note that 755 children were incorrectly identified as adults in 2025 already, while campaigners such as Foxglove and Petra Molnar of Migration and Tech Monitor compare the approach to discredited pseudoscience and past failed AI border-screening trials in Europe. The Home Office says the AI will be used alongside checks by immigration officers to make assessments "more robust and consistent".
In short
- Charities warn AI age-estimation tech will misclassify child refugees as adults
- System reportedly shows racial bias against Black children from conflict zones
- 755 children already wrongly identified as adults in 2025, Home Office data shows
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Supporters of introducing AI-assisted age assessment argue that the current system, which relies heavily on visual judgement and paperwork that is often absent or unreliable, is itself inconsistent and open to abuse, including cases where adults have been wrongly housed in child-only accommodation with vulnerable minors. They contend that a standardised, technology-assisted tool could offer a more consistent first check than variable human judgement, freeing caseworkers to focus on borderline cases, and that if bias is found it can be measured and corrected over time rather than abandoning automation altogether. From this view, robust age verification also protects genuine child refugees by ensuring safeguarding resources and child-specific placements are reserved for those who are actually children.
The case against
Critics, including the charity and human rights groups cited, argue that deploying facial-recognition age estimation now is premature and dangerous given documented evidence that such models perform less accurately on people of colour, meaning genuine child refugees from certain ethnic backgrounds are disproportionately likely to be misclassified as adults. They emphasise that the consequences of error are severe and asymmetric: a wrongly aged child placed in adult accommodation faces serious risks of exploitation, abuse and psychological harm, whereas the cost of additional manual verification for a wrongly aged adult is comparatively minor. On this view, the precautionary principle and the state's duty of care towards unaccompanied minors demand that flawed or unproven technology should not be rolled out until independent testing demonstrates it is free from racial bias.
Explained simply
The UK government plans to use artificial intelligence to assess the age of migrants arriving at the border. The system analyses facial features to estimate whether someone is an adult or a child.
Children's charities and human rights groups warn that the system can misclassify young people. They argue it tends to overestimate the age of Black children from conflict zones, which could result in minors being wrongly identified as adults and losing legal protections. This could expose them to detention, removal from the UK, or housing with adult migrants.
Research has found the technology shows racial bias, particularly against people from sub-Saharan Africa. The Home Office acknowledges that even the best age-estimation systems have a margin of error of up to 30 months—roughly two and a half years.
AI Government Politics Software Technology UK World