AI Summaries Deployed to Tackle UK Asylum Backlog Face Accuracy Concerns

In April, the Home Office introduced the Asylum Case Summarisation (ACS) tool, which uses OpenAI’s GPT-4 model to generate concise summaries of interview transcripts that can run up to 50 pages. The aim was to speed processing of the growing backlog of asylum claims. By the end of July, the Home Office was producing roughly 7,000 AI-generated summaries each month, though decision-makers could choose whether to rely on them.

The Independent

Internal spot-checks between April and July involved 259 reviews by subject-matter experts. Of the 203 primary reviews, 45 (22 %) failed accuracy standards; 19 were flagged for further review. Among 56 secondary reviews, 30 (54 %) were deemed sub-standard. Errors were defined as “multiple factual errors or misrepresentations” or “largely factually inconsistent” with the interviewee’s statements. The Home Office’s October 2025 impact assessment had projected an average of 17,515 cases processed monthly with the AI tool, far above the current output.

The Independent

It added that performance is “regularly monitored” alongside existing quality-assurance processes.

The Independent

• “It has taken eleven parliamentary questions, two ministerial letters and eight FOI requests to uncover just some information about how AI is being used in the asylum system. There are still many contradictions and crucial information remains unpublished.” — Nick Beales

The Independent

• “We are therefore concerned about flawed interview records being fed into AI software and producing unreliable decisions that do not account for the real-life circumstances in which these interviews take place.” — Mr Beales

The Independent