Politics · United Kingdom · 1 hr ago

Home Office asylum AI summaries fail 22% of initial accuracy checks

Home Office asylum AI summaries fail 22% of initial accuracy checks

A rights group says Home Office AI summaries often contain errors in the UK’s asylum system.

The tools help officials summarise asylum interviews and search country policy information, and both use OpenAI’s GPT-4o model through Microsoft’s Azure service.

Experts rejected 45 of 203 summaries checked after the summarising tool began operating in April 2026, a 22% failure rate.

The group says checks covered only an estimated 0.3% of summaries from April to July, while testing of the separate policy search tool had stopped by August.

Applicants are not specifically told when AI is used and cannot inspect or correct the summaries of their interviews.

The group also found no clear evidence of faster decisions: the average monthly number of initial decisions fell about 42% compared with the same period in 2025, while average interview-to-decision time rose 29%, though this does not prove AI caused the changes.

The group wants both tools suspended and independently audited because it says current safeguards are inadequate.

Sources

Related news