AI age checks for asylum seekers: the promise, the peril, and the people in the middle

From next year, the United Kingdom plans to do something that would have sounded like science fiction a decade ago: use artificial intelligence to look at the face of a person seeking asylum and decide, on the balance of pixels, whether they are a child or an adult.

According to the BBC, the Home Office will roll out facial age estimation technology in 2026 as part of efforts to determine how old asylum seekers are when they arrive without verifiable documents. Officials argue the tool will speed up decisions and protect genuine children from being housed with adults. Critics say it risks doing the opposite — turning a deeply human judgement about a vulnerable person into a probabilistic guess generated by a machine.

For Australian readers, this isn't just a story about Britain. It's a preview of the next frontier of border technology, one that governments around the world — including ours — will soon be asked to consider.

What the UK is actually proposing

The plan, as reported by the BBC and outlets including the Baltic News Network and Khaama Press, is to use AI-based facial analysis to estimate the age of asylum seekers whose age is disputed. The technology is expected to be deployed from 2026, sitting alongside existing methods such as interviews and document checks.

The political driver is clear. Age disputes have become a flashpoint in the UK's asylum system. If a claimant is judged to be a child, they are entitled to additional safeguarding, different accommodation, and a different legal pathway. If they are judged to be an adult, they can be placed in adult facilities and processed through the standard system. Get it wrong in either direction and the consequences are serious: a child placed among adults faces obvious safeguarding risks, while an adult wrongly classified as a child can end up in schools or foster placements.

The Home Office frames AI as a way to cut through that uncertainty. As coverage from Modern Ghana and Khaama Press notes, the technology will use facial scans to produce an age estimate that officials can weigh against other evidence.

The accuracy problem nobody can wave away

Facial age estimation is a real and improving field. Commercial systems can sort faces into broad age brackets with reasonable accuracy when the subject is a well-lit, front-facing adult in their 30s or 40s. The problem is that none of those conditions apply to the people this system is designed for.

The most important asylum age disputes cluster around a single threshold: is this person 17, or are they 18? That is precisely the band where facial age estimation performs worst. Adolescent faces change rapidly and unevenly. Puberty, nutrition, stress, sleep deprivation, sun exposure and trauma all leave marks. A 16-year-old who has spent months on a dangerous migration route may look significantly older than a 16-year-old in a school photo.

There is also a well-documented issue of demographic bias in facial analysis systems. Multiple independent audits over the past decade have shown that commercial face algorithms tend to be less accurate on darker-skinned faces and on women, partly because the datasets used to train them have historically over-represented lighter-skinned men. Asylum seekers arriving in the UK come overwhelmingly from regions — the Middle East, Africa, South and Central Asia — that are under-represented in the training data that powers most commercial systems.

In other words, the population the tool is being aimed at is the population on which similar tools have historically performed worst.

What 'AI-assisted' really means in practice

Officials are careful to describe the technology as one input among many. Caseworkers, the argument goes, will still make the final call. But anyone who has watched algorithmic tools move into public administration knows how that story tends to unfold.

Once a number is on the page — “estimated age: 19.4” — it exerts a gravitational pull on every subsequent decision. Overworked staff under pressure to clear backlogs tend to defer to the machine, especially when overruling it requires extra paperwork to justify. This is sometimes called “automation bias”, and it has been observed in everything from medical diagnostics to welfare fraud detection.

There is also the question of how a young person is supposed to challenge the algorithm. If a caseworker says “you look 20 to me”, that's a human judgement that can be argued against. If a computer says “our model places you in the 18–22 range with 84% confidence”, the asylum seeker — often without a lawyer, often in their second or third language — is being asked to contest a black box.

The ethical stakes for vulnerable people

Asylum seekers, by definition, are at one of the most vulnerable moments of their lives. Many unaccompanied minors have travelled thousands of kilometres without family. They may have lost identity documents, had them confiscated, or never possessed them. They may have been coached on what to say, or be too traumatised to say anything coherent at all.

This is the context into which the UK is introducing facial age estimation. The promise — faster decisions, less reliance on uncomfortable physical examinations such as dental X-rays — is real. Existing age assessment methods have their own serious problems. Bone and dental analyses have been criticised by paediatricians and medical bodies for producing wide error margins and exposing children to radiation for non-medical reasons. Holistic interview-based assessments are slow, subjective and inconsistent between local authorities.

Against that backdrop, an AI tool that is fast, cheap and consistent is genuinely attractive. The question is whether “consistent” is the same as “accurate”, and whether the cost of being wrong falls on the institution or on the child.

Why Australians should be watching

Australia's asylum system is structured differently from Britain's, but the underlying pressures are similar: political demand for faster processing, contested age assessments for unaccompanied minors, and a steady drift toward digital identity tools at the border. The technology being trialled in the UK from 2026 will not stay in the UK. Vendors will pitch it across the Five Eyes and the EU. Australian policymakers will at some point be asked whether to adopt, adapt or reject it.

The right questions to ask are not really about the AI. They are about the framework around it. What is the published error rate, broken down by ethnicity, gender and age band? Who audits the system, and how often? What is the appeal process when a young person disputes the result? Is the underlying model open to independent inspection, or is it a commercial secret? Are results used as one input among many, or do they become the default the caseworker has to argue against?

If those questions can be answered convincingly, AI age estimation might genuinely improve a deeply imperfect process. If they can't, the technology will simply launder old uncertainties through a new and harder-to-challenge medium.

The bottom line

The UK's 2026 rollout will be one of the largest real-world tests of facial age estimation on a vulnerable population anywhere in the world. It will generate data, court cases and almost certainly scandals. Other governments — Australia's included — will learn from what happens.

For now, the most honest framing is this: AI age estimation is neither the silver bullet its backers suggest nor the dystopia its loudest critics fear. It is a tool with real capability and real limits, being introduced at the sharpest end of public policy, where the people affected have the least power to push back. That asymmetry, more than the algorithm itself, is what deserves close attention.

Related on Bleen

Sources