UK Home Office Faces Potential Investigation Over OpenAI Asylum Tool's 22 Percent Failure Rate

The UK Home Office's artificial intelligence system for summarizing asylum interviews failed accuracy checks at twice the rate of its pilot phase, prompting the nation's immigration watchdog to seek details about the tool's deployment. A six-month investigation by the Open Rights Group revealed that the Asylum Case Summarisation tool, powered by OpenAI's GPT-4o model via Microsoft Azure, produced flawed summaries in 45 out of 203 expert-reviewed cases—a 22 percent failure rate that far exceeded the 9 percent rejection rate observed during testing. The tool was rolled out across all caseworkers in April 2026.