Home Office use of AI in asylum decisions likely unlawful, legal opinion finds

TL;DR: An 84-page legal opinion published by the Open Rights Group concludes that the Home Office’s use of AI in asylum decision-making is likely unlawful. Asylum seekers are not told AI tools are being used in their cases, and the tools have documented accuracy problems. The opinion opens the door to legal challenges.

The opinion, written by Robin Allen KC and Dee Masters of Cloisters Chambers with Joshua Jackson of Doughty Street Chambers, identifies two generative AI tools currently used in the asylum process. The Asylum Case Summarisation (ACS) tool summarises applicant interview information, while the Asylum Policy Search (APS) tool searches country-of-origin data for caseworkers.

What the tools actually do

Both tools generate new text rather than simply indexing existing information. They “funnel, filter and regurgitate important facts,” the opinion states, meaning they can omit material that a caseworker might otherwise consider. Applicants are not shown the AI-generated output and are not even informed that AI is being used.

An internal pilot found the ACS tool produced inaccurate summaries 9% of the time. Five percent of APS users were not confident in its accuracy. The opinion notes a lack of publicly available information about how accuracy has been measured or what safeguards exist to prevent errors affecting decisions.

The legal problems

The opinion identifies several potential breaches. On procedural fairness, the authors argue that asylum seekers have a common law right to know when AI is used in their cases and to see the AI-generated material, particularly the interview summaries they are well-placed to correct.

On the Tameside duty of inquiry, the Home Office may be failing to properly assess tool accuracy, the risk of bias, and whether non-AI alternatives could achieve similar efficiency gains. The opinion also flags potential UK GDPR breaches, since the ACS processes sensitive personal data including race, religion, political beliefs, and sexual orientation. No published Equality Impact Assessment exists.

Looking forward

The opinion explicitly states it opens the way to legal challenges. Given the gravity of asylum decisions and the documented accuracy failings, claims are likely. For UK organisations deploying AI in high-stakes decisions, this opinion is a warning: transparency is not optional, and “the computer says no” is not a defensible position when fundamental rights are at stake.