An investigation by The Independent reveals that the Asylum Case Summarisation (ACS) AI tool—used since April to draft case digests from interview transcripts—failed to meet basic accuracy standards in 22 % of Home Office spot-checks. A second-level audit found more than half of the reviewed summaries were ‘sub-standard’. The ACS system, built on OpenAI’s GPT-4 model, was meant to cut processing time by 23 minutes per file and help clear the asylum backlog. MPs and charities now warn that erroneous summaries could lead to wrongful refusals and additional appeals, undermining the very efficiencies the government seeks. The Home Office insists human decision-makers remain responsible and that AI outputs are supplementary, but internal documents obtained under FOI show that staff often accept the AI draft without cross-checking full transcripts. Implications for mobility stakeholders While asylum processing is distinct from work-migration routes, the story underscores the Home Office’s growing reliance on AI across immigration functions—including ETA risk-assessment and sponsorship compliance analytics. Employers should anticipate similar tools being deployed in worker-visa contexts and demand transparency on error-rates and oversight mechanisms.
Source: The Independent
