The Home Office has announced it will deploy an AI tool to estimate the age of asylum seekers starting next year, finally solving the problem of adults who look young enough to fool a system that previously relied on civil servants squinting at their faces and making a decision.

The stated goal is to catch migrants “attempting to game the system.” The system being gamed is, of course, a system that was already gamed by humans with clipboards and the confidence of someone who once read a book about phrenology.

Let us be clear about what just happened here. The Home Office looked at its existing age assessment process—which involves trained social workers conducting interviews, reviewing documents, and applying professional judgment—and decided the problem was not enough uncertainty. The solution: machine learning on photographs. A technology that has famously never made a mistake about human characteristics, especially not along demographic lines.

Why would they do this? Because the Home Office has spent the last decade proving that if a problem can be made worse with technology, it absolutely should be. The Rwanda scheme didn’t work, so they added AI. Facial recognition didn’t solve crime, so they added more data. Why would age estimation be different?

The AI tool will analyze photographs to determine whether someone is an adult trying to pose as a child, which is the kind of fraud that definitely happens at a rate that justifies replacing a system with a machine that cannot ask follow-up questions, cannot interpret context, and cannot be held accountable when it gets it catastrophically wrong.

Consider the actual mechanics. A photograph captures a moment. Faces vary wildly by genetics, nutrition, sun exposure, and whether someone has had a difficult year. The AI will look at a photo and produce a number. That number will be treated as fact. A 17-year-old who looks 19 will be classified as an adult. A 22-year-old who looks young will be classified as a child. Either outcome ruins a person’s asylum claim, which is the entire point, which is why the Home Office is doing this.

The Home Office claims the tool will “make it easier” to identify fraud. Easier than what? Easier than the current system where they already reject most child asylum seekers and reclassify them as adults anyway? The system is not broken because it is too lenient. It is broken because it is too harsh, and the Home Office wants it harder.

This is not actually about accuracy. If it were, they would spend money on trained assessors. This is about speed, deniability, and the appearance of technological objectivity. When the AI says someone is 24, no one can be blamed. The algorithm decided. The algorithm is neutral. The algorithm definitely does not reflect the biases in its training data, the preferences of the people who built it, or the political pressure to reject more claims faster.

The real absurdity is not that the Home Office is trying this. The real absurdity is that they are announcing it as progress.

They will deploy this tool. It will make mistakes. Some of those mistakes will harm children. The Home Office will say the tool is being “refined.” A year later, they will announce AI is now being used for something else, and this tool will still be running in the background, making decisions about people’s lives based on a photograph and a neural network trained on God knows what.

This is what happens when a government agency decides that the problem with a broken system is that it is not broken enough, and technology is the solution to problems created by technology and policy.