WIRED reported on October 27, 2026, that the UK’s Independent Chief Inspector of Borders and Immigration (ICIBI) requested information from Open Rights Group (ORG) while preparing the next year’s inspection priorities. The reported step was an information request, not the opening of a new investigation.
What the Home Office tools do
The Home Office’s Asylum Case Summarisation (ACS) tool summarizes asylum interview transcripts. Its Asylum Policy Search (APS) tool finds and summarizes country-policy information, including Country Policy Information Notes (CPINs) and Country of Origin Information Requests (COIRs). Decision-makers using APS still need to consult the full source documents.
The Home Office says trained caseworkers make asylum decisions and that the tools cannot decide claims in isolation. The government’s 2024 pilot evaluation describes both tools as aids that do not replace a decision-making step. WIRED also reported that documents obtained by ORG identified OpenAI’s 2023 GPT-4 and Microsoft Foundry as technologies behind the tools. It reported that asylum applicants were not told when the tools were used in their cases.
What the 2024 pilot measured
In the Home Office pilot, technical specialists reviewed ACS summaries before use. They judged 9% of the summaries inaccurate or incomplete and removed them; the affected cases went through the usual process. That figure measures summaries screened out before use, not errors in summaries used to decide asylum claims.
The evaluation found different average time savings for the two tasks:
| Tool and task | Average time difference per case | Pilot comparison |
| ACS transcript review | 23 minutes | The test group reviewed transcripts faster than the comparison group. |
| APS country-policy research | 37 minutes | The test group spent less time researching than the comparison group. |
The figures come from separate task comparisons, not a single result for every case. The pilots were small, and the participating decision-making units were chosen for availability rather than to represent all units. The evaluation found similar proportions of sustained decisions in test and comparison groups, with no apparent positive or negative effect on decision quality during the pilots. It called for larger-scale evaluation of any future rollout.
User feedback offered a separate view of the tools. Among ACS summaries that passed screening, 23% of users said they were not fully confident in the summaries and that the issue warranted further exploration. For APS, users said 82% of queries returned sufficient information and references; 79% of survey respondents were confident in its representation of a CPIN, while 5% were not. These were user assessments, not independent accuracy measurements.
The Home Office’s 2024 evaluation reports the pilot methods, screening results and task-specific time comparisons. ORG’s separate assessment of ACS quality, as reported by WIRED, judged a significantly larger share unfit for use than the Home Office’s published 9% screening result.
