According to an Open Rights Group briefing and a SOAS policy note, the UK Home Office is using two AI tools in asylum casework: the Asylum Case Summarisation (ACS) tool and the Asylum Policy Search (APS) tool. The Open Rights Group reports that both tools use GPT-4o; APS was fully rolled out in July 2025 and ACS followed in April 2026. Open Rights Group and a commissioned legal opinion concluded in March 2026 that the tools are likely unlawful, per the briefing. The brief alleges multiple governance gaps: applicants are not informed AI is used, no Data Protection Impact Assessment or Equality Impact Assessment has been published, the tools do not appear in the Algorithmic Transparency Recording Standard repository, prompts have been withheld, training and oversight are limited, and caseworkers are not required to verify AI outputs, according to Open Rights Group and SOAS.
What happened
According to the Open Rights Group briefing and the SOAS ICOP policy brief by Sara Alsherif and Sanjana Deen, the UK Home Office is using two AI tools in asylum decision-making, the Asylum Case Summarisation (ACS) tool and the Asylum Policy Search (APS) tool. Per Open Rights Group, both tools run on GPT-4o; APS was fully rolled out in July 2025 and ACS followed in April 2026. The Open Rights Group cites a commissioned legal opinion published in March 2026 that concluded the tools are likely unlawful, and it states that the Home Office subsequently confirmed it did not follow the UK AI Playbook principles, per the papers.
What the sources document
Open Rights Group and the SOAS briefing report that both tools generate new text rather than purely retrieving source documents, which allows outputs to omit or rephrase facts relevant to caseworkers. The Home Office's own pilot evaluation of ACS found that 9% of AI-generated summaries were so flawed they had to be removed, per the Open Rights Group report. The brief lists governance and safeguards gaps: applicants are not told AI is being used; no Data Protection Impact Assessment (DPIA), no Equality Impact Assessment (EIA), and no published Standard Operating Procedure (SOP) are available for either tool; neither tool appears in the Algorithmic Transparency Recording Standard (ATRS) repository; the Home Office has refused to disclose the ACS prompt; training cited by ministers is described as generic; and oversight is described as limited and reactive, relying on a later-established feedback inbox.
Technical and legal context
The Cloisters chambers legal opinion - authored by Robin Allen KC, Dee Masters, and Joshua Jackson of Doughty Street Chambers - identified breaches of procedural fairness, data protection law, and equality law. Independently, the Electronic Immigration Network has reported that the lawfulness challenge is being actively pursued. For practitioners, the case demonstrates how governance gaps compound when generative LLMs replace document retrieval in high-stakes regulated workflows: auditability, prompt transparency, DPIA compliance, and human-in-the-loop oversight each represent a distinct failure point.
What to watch
Regulatory and legal scrutiny of AI tools in public-sector decision-making is intensifying. Practitioners deploying LLMs in regulated domains should monitor how UK courts and the ICO respond to the Home Office case, as findings may set precedent for data protection and procedural fairness requirements across other jurisdictions.
Scoring Rationale #
This story affects how practitioners and auditors think about deploying generative LLMs in regulated, high-stakes public-sector decisions. The legal and governance questions documented by Open Rights Group raise compliance and audit priorities for ML teams working with public authorities.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.