UK Home Office AI tools face scrutiny over asylum decision accuracy

An investigation by the Open Rights Group (ORG) has raised concerns regarding the use of generative AI tools by the UK Home Office in asylum decision-making processes. The report focuses on two systems: the Asylum Case Summarisation (ACS) tool, which summarizes interviews, and the Asylum Policy Search (APS) tool, which assists officials in searching country policy information. Both tools utilize OpenAI’s GPT-4o model via Microsoft’s Azure AI Foundry.

Findings indicate significant accuracy issues, with one in five interview summaries failing expert checks. Specifically, 45 out of 203 summaries checked after the April 2026 rollout failed initial accuracy tests, representing a 22 percent failure rate—a notable increase from the 9 percent failure rate recorded during pilot testing. Errors identified include factual inaccuracies and misrepresentations of asylum seeker statements.

The investigation also highlights a lack of oversight, noting that only approximately 0.3 percent of generated summaries were checked by Home Office subject matter experts between April and July. Furthermore, the report claims there is no quality assurance for ACS summaries, no independent audit of the system, and that asylum claimants are not informed when AI is used to summarize their cases or country information.