UK Watchdog Weighs Probe Into Home Office's Use of OpenAI Tools on Asylum Cases

By James Calloway

The UK's borders watchdog is weighing an investigation into the Home Office's use of generative AI to process asylum applications, after researchers found error rates significantly above the government's published figures. The Independent Chief Inspector of Borders and Immigration (ICIBI) — the body that monitors the Home Office's management of borders, immigration, and customs — has asked digital rights nonprofit Open Rights Group (ORG) to share more information on the tools as it drafts its inspection priorities for next year, campaigner Sara Alsherif told WIRED.

The stakes are concrete. The tools in question summarize asylum interviews and country-of-origin information — material that feeds decisions on whether to return people to countries where they fear persecution. The ICIBI told WIRED it cannot confirm what topics are under active consideration before publishing its annual inspection priorities, but said it is aware of concerns and pointed to its own finding last year that one tool's "effectiveness and reliability were varied." A potential probe would put the UK's fastest-growing AI deployment in immigration under independent scrutiny for the first time.

What tools are involved — and what do they actually do?

The Home Office runs two systems: Asylum Case Summarisation (ACS) and Asylum Policy Search (APS). Both are built on OpenAI's 2023 GPT-4 large language model, a text-generation system, and on Microsoft Foundry, according to documents obtained by ORG. Caseworkers use the ACS to speed up review of interview transcripts, where asylum seekers describe persecution and explain why they fear returning home. They use the APS as a search assistant to find and summarize country policy information — the material that determines whether an applicant faces a genuine risk of harm.

The department began piloting the tools in late 2024 as the Labour government tried to clear a backlog of roughly 100,000 asylum cases while absorbing tens of thousands of new applications each year. Government estimates from the pilot show the two tools used together save around one hour per case.

The results so far are mixed. The department halved the number of people waiting on an initial decision in the year through June. Over the same period, the number of appeals against decisions has ballooned. Which raises the question the government has not answered: whether the two trends are connected.

How inaccurate are the AI summaries?

ORG's findings come from a six-month investigation involving 11 parliamentary questions, two ministerial letters, and 10 public records requests — all attempts to pry details out of a department that has withheld them.

What the group found is blunt. The percentage of ACS-generated summaries deemed not fit for use — often because they contain factual errors or misrepresentations — is significantly higher than the published 9 percent error rate calculated in the pilot evaluation. The APS has no quality control for its summaries at all, ORG found. The Home Office has promised updated evaluations on the rollout; it has not yet published them.

A July court case sharpened the concern. A UK judge accused the Home Office of citing an "AI hallucinated" document — fabricated content produced by a language model — to refuse a Moroccan woman and her child's asylum claim on the basis that her home country would be safe for her return.

"We're dealing with people's lives," Labour backbench MP Bell Ribeiro-Addy told WIRED. "We're talking about people that have faced some of the worst circumstances, and we're meant to be making decisions based on facts and on understanding harm. I just don't think it's appropriate that machines are doing that."

What does the Home Office say?

A Home Office spokesperson said in an email: "Asylum decisions are made by trained caseworkers, with AI tools used only to assist in navigating interview records and policy information. They cannot be used in isolation and do not make decisions on claims." The spokesperson added that the department "regularly monitor[s] the performance of these tools alongside our established quality-assurance processes, and have already committed to publishing updated evaluations on their roll out."

A second spokesperson statement stressed: "Use of Artificial Intelligence is just one of the tools at our disposal to protect our borders, and does not replace or automate the role of human decision-makers."

The caveat is what those statements omit. The department says caseworkers remain in the loop; it does not say how much the hour-per-case time saving depends on trusting AI summaries, nor why the observed error rate diverges from the published one. An OpenAI spokesperson pointed to the company's ongoing work to improve model accuracy and noted it will retire GPT-4 in January — which means the Home Office's tools may soon run on a model whose successor has not been evaluated in this context.

Is the government breaking its own transparency rules?

The UK has no AI regulation on the statute books for this. In its place, the government's "AI Playbook" mandates that departments be transparent about algorithmic tools used in decision-making, make information about them public, and list them on a public repository. The Home Office's case summarization and policy search tools are not on that list.

Asylum seekers are also not told when the tools are used on their case. "Migrant organizations, legal practitioners, not many of them know about the actual use of this tool," said Alsherif, who leads ORG's migrant digital justice program. "So we need to let them know about this, let them know how it works, so they can challenge it." Lawyers at Cloisters wrote in an independent legal opinion commissioned by ORG that being unable to contest an incorrect AI summary of one's personal information may breach UK data rights laws. The Home Office's privacy notice states only that parts of the decision-making process may involve automation.

There is a bias concern as well. Ribeiro-Addy, Alsherif, and the Public Law Project, which has separately investigated the tools, point to well-documented "cultural bias" in GPT models that could flatten language barriers and cultural nuances in summaries — potentially producing discriminatory decisions.

How much AI is government running in secret?

Both ORG and the Public Law Project have sent their findings to the Information Commissioner's Office, the UK data regulator, which is currently investigating the Home Office's refusal to tell ORG how staff use, and were trained to use, the two tools.

The wider picture may be the most troubling finding. Public Law Project senior research fellow Caroline Selman said the ACS and APS are just two of 11 high-risk AI systems her team has found in use across government but absent from the public repository.

"Other tools are being used but not being publicized at all," Selman said, "and you can't even begin to ask questions about things if you don't know that they're being used." She fears there are more deployments that even researchers — let alone the people they are used on — cannot see.

The ICIBI's decision on next year's inspection priorities, the ICO's ongoing investigation, and the promised but unpublished evaluations will together determine whether the Home Office's asylum AI faces genuine accountability or continues to operate on trust.