• Home
• Business

A new report by rights group Open Rights Group (ORG) has raised concerns over the UK Home Office's use of artificial intelligence in asylum processing, finding that 22% of AI-generated interview summaries examined by experts failed initial accuracy checks.

The 20-page investigation, based on Freedom of Information requests and correspondence with ministers, questions the reliability, oversight and promised efficiency gains of two AI tools used by Home Office caseworkers.

The systems are Asylum Case Summarisation (ACS), which produces summaries of asylum interviews, and Asylum Policy Search (APS), a chatbot-style tool for searching country policy information. Both run on OpenAI's GPT-4o model through Microsoft's Azure AI Foundry.

Accuracy concerns over AI-generated summaries

Data obtained by ORG show that 45 of 203 AI-generated interview summaries failed initial expert checks after ACS was rolled out in April 2026. That represents a 22% failure rate, up from the 9% rejection rate recorded during the original pilot.

A further 19 summaries were flagged for additional review. The figures also list 56 second-stage reviews, of which 30 failed to meet the required standard.

Errors can include factual inaccuracies and misrepresentations of statements made by asylum seekers during interviews. ORG also found that summaries could pass checks despite leaving out important details, provided the information they retained was accurate.

The group estimates that Home Office experts reviewed just 0.3% of summaries between April and July, using the department's projected monthly caseload of 17,515 asylum cases as a benchmark.

Questions over efficiency and human oversight

ORG also challenged claims that the tools are accelerating asylum decisions. It reported that the average monthly number of initial substantive asylum decisions fell by around 42% from April 2025 to July 2025, compared with the corresponding period in 2026. Over the same comparison, the mean time between interview and decision increased by 29%.

These figures do not establish that AI caused the slowdown, but they provide little evidence of the productivity improvements promised by ministers.

Oversight of APS has also come under scrutiny. In August, the Home Office confirmed that expert testing had stopped, despite previously telling MPs that checks were ongoing. ORG said no summary of the test results had been produced.

Asylum applicants are not specifically informed when AI is used in their cases and cannot access or correct the summaries generated from their interviews.

The group also questioned the extent of human supervision. The Home Office's own data protection assessment says ACS automates work previously carried out by technical specialists, who reviewed and summarised asylum cases before passing them to decision-makers. ORG warned that the system could eliminate an experienced layer of scrutiny rather than simply reduce workloads.

The organisation is calling for both tools to be suspended and independently audited, arguing that existing safeguards are inadequate for decisions affecting whether people can remain in the UK.

The criticism follows earlier concerns over the Home Office's technology plans. In June, more than 60 rights groups urged ministers to abandon plans for AI-powered age estimation for asylum-seeking children, warning that the technology could be unreliable when distinguishing people aged 16 to 18.

Later that month, an inspection found that the department's long-delayed Atlas immigration system was failing to provide caseworkers with feedback from asylum appeals, leaving staff to rely on workarounds.