ORG uncovers AI governance failings in two Home Office tools used in asylum decisions

A six-month investigation by Open Rights Group (ORG), supported by MPs Bell Ribeiro-Addy and Carla Denyer, found that the Home Office is using two generative artificial intelligence (AI) tools in all asylum decisions without adequate testing or safeguards.

The systems, built using OpenAI's GPT-4o via Azure AI Foundry, generate new text that can reframe or omit critical facts. ORG rolled out the Asylum Policy Search (APS) tool to all decision-makers in July 2025, followed by the Asylum Case Summarisation (ACS) tool for all caseworkers in April 2026.

ORG's investigation found that asylum claimants are not notified when AI processes their cases, which the group argues is likely unlawful. Furthermore, pilot evaluations and sample testing revealed error rates of 9% and 15%, respectively, rendering their outputs unfit for purpose. Despite this, neither tool has built-in quality assurance measures, and subject matter experts reviewed only 0.3% of summaries. The department, which has refused to disclose prompt designs or training details, has not registered the systems on the Algorithmic Transparency Recording Standard Hub nor aligned them with the central government AI Playbook.

Training Announcement: The BCS Foundation Certificate in AI examines the challenges and risks associated with AI projects, such as those related to privacy, transparency and potential biases in algorithms that could lead to unintended consequences. Explore the role of data, effective risk management strategies, compliance requirements, and ongoing governance of the AI lifecycle and become a certified AI Governance professional. Find out more.