Glonce

UK watchdog weighs probe of Home Office's GPT-4 asylum tools

policyThe Glonce editorsWired🔥56Glonce rating56Impact55Novelty40Relevance80Surprise40Players35Credibility72
UK watchdog weighs probe of Home Office's GPT-4 asylum tools

A UK government watchdog is considering an investigation into the Home Office's use of generative AI tools to process asylum applications, Wired reports. The Independent Chief Inspector of Borders and Immigration (ICIBI), which monitors the Home Office's management of borders, immigration and customs, has asked the digital rights nonprofit Open Rights Group (ORG) for more information on the tools as it draws up the topics it will inspect next year. That is according to ORG campaigner Sara Alsherif. The ICIBI itself told Wired it cannot confirm which topics are under active consideration before it publishes its annual inspection priorities. It said it is aware of concerns about AI tools and pointed to its findings last year that one tool's 'effectiveness and reliability were varied'.

The two tools are Asylum Case Summarisation (ACS) and Asylum Policy Search (APS). According to documents obtained by ORG, they are built on OpenAI's 2023 GPT-4 model and Microsoft Foundry. Caseworkers use ACS to speed up their review of transcripts of interviews in which asylum seekers explain how they were persecuted and why they fear returning. They use APS as a search assistant that finds and summarises country policy information, which informs whether an applicant faces a genuine risk of harm at home. The Home Office began piloting them in late 2024, as the Labour government tried to clear a backlog of around 100,000 asylum cases. Government estimates say that used together, the tools could save around an hour per case. The department has halved the number of people waiting on an initial decision in the year through June, but the number of appeals against decisions has ballooned.

ORG's findings come from a six-month investigation that used 11 parliamentary questions, two ministerial letters and 10 public records requests to try to get details from a Home Office that has withheld them. ORG found that the share of ACS summaries judged not fit for use, often because of factual errors or misrepresentations, is significantly higher than the 9 percent error rate published from the pilot evaluation. It also found no quality control for APS summaries. In July, a UK judge accused the Home Office of citing an 'AI hallucinated' document to refuse the asylum claim of a Moroccan woman and her child, on the basis that Morocco would be safe to return to.

Labour backbench MP Bell Ribeiro-Addy told Wired: 'We're dealing with people's lives.' She added that she does not think it appropriate for machines to do this work. Ribeiro-Addy, Alsherif and the Public Law Project, which has separately investigated the tools, also worry that well-documented cultural bias in GPT models could cause language barriers and loss of cultural nuance in summaries, and in turn discriminatory decisions.

The transparency complaints are separate. The UK's AI Playbook says departments must be transparent about algorithmic tools used in decision-making and list them on a public repository. ACS and APS are not on it. Asylum seekers are not told when the tools are used in their case. Cloisters lawyers, in an independent legal opinion commissioned by ORG, wrote that being unable to contest an incorrect AI summary of their personal information may breach UK data rights laws. The Home Office says in its privacy notice that parts of the decision-making process may involve automation. ORG and the Public Law Project have sent their findings to the data regulator, the Information Commission, which is currently investigating the Home Office's refusal to tell ORG how staff use the two tools and were trained to use them. Public Law Project senior research fellow Caroline Selman said ACS and APS are two of 11 high-risk AI systems her team has found in use across government but absent from the public repository.

The Home Office rejects the criticism. A spokesperson said asylum decisions are made by trained caseworkers, with AI used only to help navigate interview records and policy information; the tools cannot be used in isolation and do not make decisions on claims. The department says it monitors performance alongside its quality-assurance processes and has committed to publishing updated evaluations of the roll-out. An OpenAI spokesperson pointed to the company's ongoing work to improve accuracy and noted that it will retire GPT-4 in January.

Key facts

  • The ICIBI, the UK borders and immigration inspector, asked Open Rights Group for more information as it plans next year's inspection topics, but says it cannot confirm what is under consideration.
  • The Asylum Case Summarisation and Asylum Policy Search tools run on OpenAI's 2023 GPT-4 and Microsoft Foundry; the pilot began in late 2024, with government estimates of around an hour saved per case when both are used.
  • ORG, after a six-month investigation, says the share of ACS summaries deemed not fit for use is significantly higher than the published 9 percent error rate, and that APS summaries have no quality control.
  • Neither tool is on the public repository the UK's AI Playbook requires, and asylum seekers are not told when they are used; the Information Commission is investigating the Home Office's refusal to disclose how staff use them.
  • The Home Office says caseworkers make all decisions and the tools only assist; OpenAI says it will retire GPT-4 in January.

Why it matters

Asylum decisions can mean sending someone back to a country where they fear persecution, so errors in AI-generated summaries of interview transcripts and country policy carry heavy consequences. The story joins several threads: a possible official inspection, a data regulator inquiry, a judge's accusation that the Home Office cited an 'AI hallucinated' document, and the finding that the tools are missing from the public register the government's own AI Playbook requires. It is also a concrete case of a GPT-4-based system used in a high-stakes public decision process.

Who it affects

Asylum applicants come first: they are not told when the tools are used in their case, and ORG's lawyers say they may be unable to contest an incorrect summary. Home Office caseworkers and decision-makers use the tools to save time when reviewing asylum cases. Migrant organisations and legal practitioners, who Alsherif says mostly do not know about the use, need to understand it to challenge decisions. The ICIBI, the Information Commission, MPs and researchers such as ORG and the Public Law Project are all drawn in. Selman's team has found 11 high-risk AI systems across government that are absent from the public repository.

How to use it

This is a news report, not a product. The practical reading is for people who work with or on behalf of asylum applicants. ORG's Alsherif says the aim is to tell migrant organisations and legal practitioners about the tools and how they work so decisions can be challenged. The Home Office has committed to publishing updated evaluations of the roll-out, and the ICIBI will publish its annual inspection priorities; both are the places to look for further detail.

How solid is it

The reporting is by Wired and rests on several named sources: ORG's documents and findings, the ICIBI's statement, Home Office and OpenAI spokespeople, an MP and a Public Law Project researcher. The key accuracy claim is ORG's: that the share of unfit ACS summaries is significantly higher than the published 9 percent, with no exact figure given. The ICIBI has not decided or announced an investigation; it only says it cannot confirm topics before publishing its annual priorities. The Cloisters legal opinion was commissioned by ORG, and it says the lack of a right to contest 'may' breach data rights law.

Risks and caveats

The headline development is a possible probe, not a confirmed one. ORG's error comparison has no published figure, so the scale of the gap is unknown. The concern that cultural bias in GPT models could produce discriminatory decisions is a worry raised by Ribeiro-Addy, Alsherif and the Public Law Project, not a finding. The source does not say that any asylum decision was actually made wrongly because of the tools, other than the judge's accusation about one document. The Home Office stresses that humans decide and the tools cannot be used in isolation. The outcome and timescale of the Information Commission investigation are not given.

“I just don’t think it’s appropriate that machines are doing that.”

— Bell Ribeiro-Addy, Labour backbench MP, to Wired