Home Office uses ChatGPT to summarise asylum interviews – and 9% come out wrong
Key Points
- ome Office asylum summary tools run on OpenAI’s ChatGPT-4o
- 9% of AI summaries contained inaccurate or missing information
- The tools made no difference to the quality of asylum decisions
- Lawyers found the Home Office’s use of the tools likely unlawful
- Open Rights Group wants ChatGPT banned from asylum applications
The Home Office’s ChatGPT-based asylum tools produced inaccurate or incomplete summaries 9% of the time, according to Open Rights Group.
The digital rights campaign group highlighted the figures in a new report on how AI is spreading across the UK’s immigration system.
The Home Office tested two tools in pilots between May and December 2024, then published its own evaluation of the results in April 2025.
The first tool, Asylum Case Summarisation, uses AI to summarise transcripts of asylum interviews for the officials deciding each case.
The second, Asylum Policy Search, finds and summarises the country policy information that caseworkers use to judge whether applicants face danger at home.
Both tools run on OpenAI’s ChatGPT-4o, which often cannot show where its summaries came from or how it produced them. The Home Office’s evaluation found that 9% of the summaries contained inaccurate or missing information.
Almost a quarter of the officials interviewed, 23%, said they did not feel fully confident in the summaries.
The Home Office said the summary tool could save 23 minutes per case, while the policy search tool could save an average of 37 minutes.
Some caseworkers in the pilots reported minimal time savings because the summaries did not include source references.
The evaluation also found that the tools made no difference to the quality of decisions in either direction.
Legal challenge risk
A legal opinion commissioned by Open Rights Group found that the Home Office’s use of the tools was likely unlawful.
The lawyers found limited testing for bias, no published Equality Impact Assessment and no entry for either tool in the government’s algorithmic transparency register.
They also said the Home Office failed to tell asylum applicants when it used AI on their cases, and that the summary tool may breach data protection rules.
The Home Office later confirmed to Parliament that it did not follow the government’s own AI Playbook when deploying the tools.
Open Rights Group said the legal opinion opens the way for asylum applicants to challenge decisions where they believe AI played a role.
“If an asylum seeker believes that an AI system has made a mistake, they need to be able to know that the technology was used and have the chance to challenge its output,” said Pete Wishart, SNP MP.
“We should be careful not to assume that making the system faster necessarily makes it better,” said Wishart.
The Home Office went on to explore AI for interpreting, translating and transcribing speech in languages other than English in October 2025.
Open Rights Group wants the government to ban the use of ChatGPT tools in asylum applications entirely.
