The UK Home Office's AI-powered asylum interview tool failed more than one in five accuracy checks conducted by experts, according to a new investigation published by the Open Rights Group (ORG). The 20-page report released this week claims the department is moving forward with artificial intelligence systems despite questions about their precision, monitoring, and ability to deliver the efficiency improvements ministers have pledged. The findings center on two systems caseworkers use to process asylum applications, both powered by OpenAI's GPT-4o through Microsoft's Azure platform.
Freedom of Information data obtained by ORG reveals that 45 out of 203 AI-generated interview summaries failed initial expert reviews after the Asylum Case Summarisation (ACS) system launched in April 2026—a 22 percent failure rate. That compares with a 9 percent rejection rate during the original pilot phase. Another 19 summaries were marked for additional review, while 56 second-stage reviews were conducted, of which 30 didn't meet required standards. The mistakes can range from factual errors to misrepresenting what asylum seekers actually said during their interviews, and summaries could still pass checks even when they left out important details as long as the information included was accurate. Based on the department's projected monthly workload of 17,515 cases, ORG estimates experts examined just 0.3 percent of summaries between April and July.
The report also questions whether the tools are delivering time savings. The average monthly number of initial substantive asylum decisions since April 2026 dropped roughly 42 percent compared with the same period in 2025, while the mean time from interview to decision climbed 29 percent. Though these figures don't establish that AI caused the delays, they provide little proof of the productivity gains officials have promised. Oversight of the second tool, Asylum Policy Search (APS)—a chat-style system helping staff search country policy information—also appears insufficient. The Home Office confirmed in August that expert testing had stopped, despite earlier telling members of Parliament that checks were ongoing, and no summary of testing results was ever produced, ORG says.
The report challenges the Home Office's claims of human oversight, noting that the department's own data protection assessment states ACS automates work previously handled by technical specialists who reviewed and summarized asylum cases before passing them to decision-makers. ORG warns the tool could eliminate an experienced layer of scrutiny rather than simply freeing up staff time. Meanwhile, asylum applicants aren't specifically informed when AI has been used in their cases and can't inspect or correct the summaries created from their interviews. The government maintains that humans remain responsible for asylum decisions, but the campaign group questions how reliably caseworkers can identify mistakes in AI-generated material.
ORG is calling for both systems to be suspended and independently audited, arguing that current protections are inadequate for decisions that determine whether someone can stay in the UK. The report points to a pattern of technology troubles at the department, including pressure from more than 60 rights groups in June to abandon AI-powered age estimation for asylum-seeking children over reliability concerns, and an inspection the same month that found the long-delayed Atlas immigration system was failing to provide caseworkers with feedback from asylum appeals. The Home Office didn't respond to questions about the findings. For agencies racing to deploy generative tools in high-stakes environments, the tension between speed and reliability may prove harder to resolve than the technology itself.