More than one in five AI-generated asylum interview summaries failed initial expert accuracy checks after the Home Office rolled out the system in April 2026, according to a new Open Rights Group report.
The 20-page investigation draws on Freedom of Information requests and correspondence with ministers. It examines two tools used by caseworkers: Asylum Case Summarisation (ACS), which summarizes interviews, and Asylum Policy Search (APS), a chat-style tool for finding country policy information. Both use OpenAI’s GPT-4o through Microsoft’s Azure AI Foundry.
Summary checks found more errors after rollout
Figures obtained by the Open Rights Group (ORG) show that 45 of 203 ACS summaries failed initial expert checks, a 22 percent failure rate. The rejection rate in the original pilot was 9 percent. Another 19 summaries were flagged for further review. Of 56 second-stage reviews, 30 failed to meet the required standard.
Problems included factual errors and misrepresentations of what asylum seekers said. ORG found that a summary could still pass if it left out important details, provided the information it included was accurate.
ORG estimates that experts checked only 0.3 percent of summaries between April and July, using the department’s projected monthly caseload of 17,515. Applicants are not specifically told when AI is used in their cases and cannot inspect or correct summaries of their interviews.
Decision figures show no clear time savings
ORG says the average monthly number of initial substantive asylum decisions since April 2026 was about 42 percent lower than in the same period of 2025. The mean time from interview to decision rose 29 percent. Those figures do not establish that AI caused the slowdown, but they offer little evidence of the faster processing ministers promised.
The Home Office says people remain responsible for asylum decisions. ORG questions whether caseworkers can reliably catch errors in AI-generated material. The department’s data protection assessment says ACS takes on work previously done by technical specialists who reviewed and summarized cases before handing them to decision-makers. ORG says this could remove an experienced layer of scrutiny.
Read nextPro-AI super PAC network paid millions to four Nevada LLCsChecks on APS have also changed. In August, the Home Office confirmed that expert testing had stopped, despite previously telling MPs the checks were continuing. ORG says no summary of the test results was produced.
The concerns follow other criticism of Home Office technology. In June, more than 60 rights groups urged ministers to drop plans for AI age estimation of asylum-seeking children, citing unreliability around ages 16 to 18. Later that month, an inspection found that the delayed Atlas immigration system was not giving caseworkers feedback from asylum appeals, leaving staff to use workarounds.
The Home Office did not respond to The Register’s questions. ORG is calling for ACS and APS to be suspended and independently audited.
