United KingdomPublic sector2024 – ongoing

UK Home Office

Asylum Case Summarisation and Asylum Policy Search AI tools

Home Office pilots reported 23 minutes saved per case with summarisation and 37 minutes on policy research; 9% of summaries were inaccurate or incomplete

In progressCompany's own claim

Problem

Asylum decision-makers had to read long interview transcripts and search country policy documents by hand, with about 90,000 cases waiting for a first decision at the end of 2024.

Approach

The Home Office piloted two tools between May and December 2024: one summarises interview transcripts (built on OpenAI's GPT-4 per the government's AI hub), the other searches and summarises country policy information. Test and comparison groups of decision-makers worked in separate units. The department published the evaluation in April 2025.

Results

23 minutes (32%)time saved per case with the summarisation tool versus the comparison groupCompany's own claim[1]

37 minutestime saved per case on policy research with the policy search toolCompany's own claim[1]

9%of summaries judged inaccurate or incomplete and removed from the pilotCompany's own claim[1][2]

23%of users not fully confident in the summary informationCompany's own claim[1]

Takeaway

Critics argued that the 9% rate of inaccurate or incomplete summaries was too high for asylum decisions. The reported time savings came from small pilots in unrepresentative units and relied on staff reports.

Sources

Evaluation of AI trials in the asylum decision making process
GOV.UK (Home Office)29 Apr 2025Government publication
Home Office asylum AI summarisation tool critique
FoxgloveMay 2025News article

Source summaries

[1]GOV.UK (Home Office)29 Apr 2025

The Home Office evaluated two pilots. Summarisation ran in two 8-week phases in 2024 and saved 23 minutes per case (32%); policy search saved 37 minutes. 9% of summaries were removed as inaccurate, 23% of users lacked full confidence, and 65% and 54% would keep using the two tools. Quality reviews showed no effect on decisions, and the department said the pilots were small and relied on staff reports.

[2]FoxgloveMay 2025

A digital-rights group argues the tool is not ready for asylum decisions. It cites the Home Office's own finding that 9% of summaries were inaccurate or incomplete, notes that nearly a quarter of testers were not fully confident, and asks for answers on error rates and bias before deployment.

Full details are in the original sources.

Added 25 Oct 2026 · Updated 25 Oct 2026