Britain’s borders inspectorate has asked the Open Rights Group for more detail on two AI tools the Home Office uses in asylum casework, while it draws up the topics it will inspect next year.
The Independent Chief Inspector of Borders and Immigration has not committed to an inquiry. The group’s campaigner, Sara Alsherif, confirmed the request to Wired, and the inspector has previously described one tool’s reliability as varied.
Both systems run on OpenAI’s GPT-4o through Microsoft Azure. Asylum Policy Search, live for all decision-makers since July 2025, offers a chat interface over country guidance. Asylum Case Summarisation, which reached caseworkers in April 2026, condenses interview transcripts.
The accuracy numbers moved the wrong way. During the Home Office’s own pilot, 9 percent of summaries were judged inaccurate or incomplete. Post-rollout checks on 203 summaries put the failure rate at 22 percent.
Oversight is thin. Subject matter experts reviewed only 0.3 percent of the roughly 7,000 summaries produced monthly, and applicants are never told AI touched their case or shown what it produced.
Open Rights Group spent six months on the question, filing 11 parliamentary questions, 10 public records requests and two ministerial letters. It obtained a legal opinion concluding the absence of notification could make the tools unlawful, and has called for suspension.
Neither tool appears in the government’s public register of algorithmic systems. The Home Office maintains that trained caseworkers make every decision and that AI only helps navigate records and policy.
