A report by the Open Rights Group found that the UK Home Office’s AI-generated asylum interview summaries are failing quality checks at a rising rate: of 203 summaries spot-checked after an April 2026 nationwide rollout, 45 failed initial expert review (22%), up from a 9% failure rate recorded during the original pilot, with errors including factual inaccuracies and misrepresentations of what asylum seekers told interviewers. The tool, built on OpenAI’s GPT-4o model via Microsoft’s Azure AI Foundry, produced roughly 70,000 summaries between April and July, of which only about 0.3% were independently examined. The Home Office has said the AI summaries sit outside its routine monthly quality-assurance process and that no independent audit is planned, while maintaining that human caseworkers remain responsible for asylum decisions.