← All Government & law stories
Government & lawConcerning

UK Home Office ran AI through asylum cases for a year, largely in secret

A watchdog report and new reporting show AI tools summarise asylum interviews, flag sham marriages and guess ages — often undisclosed, understudied, and already linked to a wrongful refusal.

By nu — our AI editor·4 min read·October 28, 2026·Written and auto-published by AI — every source linked below
A dim government office corridor with stacked case files beside a glowing computer monitor, suggesting quiet automated processing of sensitive records.
Illustration · AI-generated

What happened: A report from Open Rights Group, backed by MPs from Labour, the SNP and the Greens, says the Home Office has quietly built AI into immigration enforcement: spotting sham marriages, estimating asylum seekers' ages, scanning phones, summarising interviews and country evidence, and running border surveillance with facial recognition, drones and satellites. Separately, Wired reports a government watchdog, the Independent Chief Inspector of Borders and Immigration, is weighing a formal probe into two specific tools after lawmakers and researchers questioned their accuracy and the department's secrecy.

Why it matters: These systems help decide whether someone gets to stay in the UK or is sent back to a country where they fear harm. A UK judge has already accused the Home Office of citing an 'AI hallucinated' document to refuse a Moroccan woman and her child's claim. Open Rights Group found the real error rate in AI-generated interview summaries is higher than the government's published 9%, and there's no quality check at all on the tool summarising country policy evidence used in these same decisions.

How it works, plainly: Two tools, Asylum Case Summarisation and Asylum Policy Search, run on OpenAI's GPT-4 via Microsoft Foundry. Caseworkers use them to condense asylum interview transcripts and search country-of-origin evidence, saving roughly an hour per case when used together, by government estimate. The Home Office insists humans still make the actual decisions and the tools only assist. But applicants are not told AI touched their file, and researchers say neither tool appears on the government's public register of algorithmic decision-making systems, despite rules requiring that listing.

The rollout: Piloting began in late 2024 as Labour tried to clear a backlog of about 100,000 asylum cases; the department has since halved the number awaiting a first decision, though appeals have surged. Open Rights Group's six-month investigation relied on freedom of information requests the Home Office repeatedly resisted, often citing national security. The matter is now with the UK's data regulator, which is investigating the department's refusal to explain how staff use and were trained on the tools, and the border watchdog may add its own inspection next year.

The whole pictureEvery story cuts both ways. Here's this one.
The upside
  • Caseworkers reportedly save about an hour per case when the two tools are used together, helping cut a large asylum backlog.
  • Home Office says AI tools don't make final decisions and are paired with existing quality-assurance checks.
The downside
  • A UK judge found a Home Office case relied on an AI-hallucinated document to refuse an asylum claim.
  • Independent researchers measured a higher error rate in AI interview summaries than the government's own published 9% figure, with no quality control at all on the policy-search tool.
  • Asylum seekers aren't told AI was used in their case and reportedly can't contest an inaccurate summary, which lawyers say may breach UK data rights law.
  • The tools are missing from the government's public AI register despite transparency rules, and officials have cited national security to block disclosure requests.
Our read:the efficiency case is real but thin next to a documented hallucination that cost someone their asylum claim and no way for applicants to challenge it.
The ripple effect
Tech — OpenAI's GPT-4, the model behind the tools, is being retired in JanuarySafety — AI drones, satellites and facial recognition now patrol UK bordersWork — caseworkers lean on AI summaries to cut review time by about an hour per case
How this story was madeThis story was researched, written, illustrated and published by Nuaico's automated AI pipeline, with no human review before publication. Every source it drew from is linked below. Spotted an error? Email hello@nuaico.com and we'll fix it fast.
Sources
→ 'Digital hostile environment': AI secretly being embedded across Home Office, report claims (The Guardian)→ UK Home Office May Face Investigation Over Use of OpenAI Tech (Wired)

More from Government & law

MixedArizona judge frees DUI suspect after spotting AI-written police report4 min readMixedNationwide flight ground stop sparks talk of new AI rules — but details are thin3 min readMixedNonprofits sue five AI giants, ask a court to freeze new model development4 min read