The Reliability Gap in Asylum Processing
The UK Home Office’s initiative to streamline asylum decision-making through artificial intelligence has encountered significant turbulence. Recent findings from the Open Rights Group (ORG) indicate that the government's AI-driven systems are falling well short of accuracy benchmarks. The investigation, which analyzed data from the months following the system's April 2026 rollout, highlights that over one in five interview summaries generated by the AI failed to pass expert quality checks. This 22 percent failure rate marks a stark increase from the 9 percent rejection rate recorded during the initial pilot phase, casting doubt on the efficacy of the deployment.
The systems in question—Asylum Case Summarisation (ACS) and Asylum Policy Search (APS)—operate using OpenAI's GPT-4o model, accessed through the Microsoft Azure AI Foundry. While the government maintains that final decisions remain firmly in human hands, critics argue that the reliance on AI for summarization introduces a new layer of potential bias and factual error that busy caseworkers may not be equipped to catch. Of the 203 summaries scrutinized by experts, 45 were deemed failures, with an additional 19 flagged for further concern, underscoring the volatility of the automated output.
The Productivity Paradox
Beyond the immediate accuracy concerns, the implementation of these AI tools has failed to deliver the promised administrative efficiency. Data shows that the mean time from interview to decision has actually increased by 29 percent compared to the same period in 2025, while the number of initial substantive asylum decisions has plummeted by 42 percent. Although correlation does not necessarily imply causation, the lack of demonstrated productivity gains suggests that the technology has not fulfilled the government's ambitious narrative of faster, more efficient processing.
Furthermore, the oversight mechanisms for these tools appear inconsistent. The Home Office admitted that expert testing for the Asylum Policy Search tool ceased in August, despite prior assurances to Parliament that rigorous monitoring was ongoing. Compounding these issues is a lack of transparency for asylum seekers, who are reportedly not informed when AI is utilized in their cases and have no mechanism to review or contest the AI-generated summaries that could influence their future.
Why it Matters
- Accountability: The shift from human-led summaries to automated ones removes a vital layer of experienced scrutiny from the asylum process.
- Transparency: Asylum seekers currently have no right to inspect the AI-generated documents that form part of their official record.
- Performance: Despite the integration of advanced LLMs, key metrics regarding processing speed have worsened, challenging the business case for the technology.
- Ethics: Rights groups are calling for an immediate suspension and independent audit of the tools, citing the profound human consequences of errors in high-stakes immigration decisions.
As the debate continues, the Home Office remains under pressure to reconcile its push for digital transformation with the ethical mandates of its public-facing responsibilities. With existing concerns over other failed immigration IT projects, the current performance of the asylum AI tools adds to a growing pattern of skepticism regarding the government's ability to safely and effectively automate sensitive public sector workflows.
