Watchdog says asylum AI fails accuracy checks
Open Rights Group says Home Office tools generate errors as checks remain limited.
The Open Rights Group says artificial intelligence tools used in Home Office asylum casework have produced inaccurate summaries and are subject to limited checks. Its findings are based on a six-month investigation of ministerial correspondence and Freedom of Information responses.
The organisation said on 12 October that its investigation examined two systems: Asylum Case Summarisation, which condenses interview transcripts, and Asylum Policy Search, which helps staff search and summarise country-policy material. Both generate new text, ORG said, meaning they can omit or reframe information rather than simply point caseworkers to an original document.
The report is not an independent inspection commissioned by the Home Office. ORG said Home Office subject-matter experts checked only 0.3% of all generated summaries between April and July. It also said the Home Office had no quality-assurance process for the summarisation tool, no quality-check process for the policy-search tool and no plan for an independent audit.
ORG compared the live use of the system with the Home Office's earlier pilot evaluation. The government's research note, published in April 2025, said 9% of summaries were inaccurate or omitted information and had to be removed before use in the pilot. ORG said later sampling showed that accuracy problems remained, but stressed that the available samples did not measure every asylum case.
According to ORG, the summarisation tool was rolled out to all asylum caseworkers in April 2026, while the policy-search tool had been fully rolled out to asylum decision-makers in July 2025. Government materials describe the summarisation system as a human-in-the-loop tool. ORG says applicants are not told when AI is used to summarise their material and cannot see or correct an inaccurate summary.
The Home Office's pilot materials said the summarisation tool saved an average of 23 minutes per case. They also recorded that 23% of caseworkers were not fully confident in its output. ORG said prompts, details of staff training and data protection impact assessments had not been published in full.
ORG is calling for both tools to be suspended, followed by an independent audit of their live accuracy and an inspection by the Independent Chief Inspector of Borders and Immigration. Those are campaign recommendations, not decisions already taken by ministers. The evidence supports scrutiny of how the systems are governed; it does not establish that AI alone determines asylum outcomes.
One story, several perspectives
What is established
- The Home Office uses AI tools to summarise asylum interviews and search country-policy material.
- The Home Office's pilot found accuracy problems, while ORG reports further concerns about live checking and transparency.
- The tools are described by government material as assisting human decision-makers rather than replacing them.
Left
Arguments High-risk immigration decisions should not rely on opaque generated text until independent testing, disclosure to applicants and correction rights are in place.
Values Due process, equality, transparency and protection for people with limited power to challenge the state.
Consequences Continuing without stronger safeguards could allow errors to influence decisions and make later legal remedies harder.
Centre
Arguments AI may help caseworkers manage a large workload, but only within a tightly governed system with human responsibility, published performance standards and regular independent audits.
Values Proportionality, institutional accountability and workable public services.
Consequences A blanket ban could slow decisions, while unrestricted deployment could undermine confidence in the asylum process.
Right
Arguments The state needs tools that improve productivity and reduce administrative delay, provided officials remain accountable for final decisions and poor performance is exposed quickly.
Values Effective government, border control and responsible use of taxpayers' money.
Consequences Removing useful technology could increase backlogs and costs, but demonstrably unreliable systems should be corrected or withdrawn.
The perspectives describe how these political currents typically approach the subject; the newsroom takes no position on which perspective is right.
Fact-check Approved · Priya Shah — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The article clearly attributes the latest allegations to Open Rights Group and uses official Home Office material for the earlier pilot. No unsupported claim that AI alone decides asylum cases is made.
- confirmed Open Rights Group published a six-month investigation on 12 October 2026. — The publication is dated 12 October 2026 and describes a six-month investigation. source
- confirmed The Home Office uses Asylum Case Summarisation and Asylum Policy Search in asylum casework. — ORG describes both tools and the government AI hub documents the summarisation system. source
- confirmed Only 0.3% of generated summaries were checked between April and July. — This figure appears in the ORG investigation. source
- confirmed The Home Office pilot found 9% of summaries inaccurate or incomplete and removed them. — The official Home Office evaluation records this result. source
- confirmed The pilot reported a 23-minute average time saving and 23% of caseworkers lacked full confidence in ACS output. — Both figures are included in official government materials. source
- confirmed ORG has called for suspension, an independent audit and an inspection. — These are listed recommendations in the investigation. source
Editor's note
The Home Office's pilot findings are official. The newer allegations about live accuracy, transparency and checking come from an Open Rights Group investigation and have not been independently audited.Sources
- The Use of ChatGPT in Asylum Decisions — Open Rights Group
- Evaluation of AI trials in the asylum decision making process — Home Office
- Asylum Case Summarisation (ACS) — UK Government AI Knowledge Hub
More on this in other media
- The Guardian — „artificial intelligence”
- BBC News — „artificial intelligence”
- Sky News — „artificial intelligence”