UK Government Watchdog Eyes Home Office Over Secretive Use of Generative AI in Asylum Processing

LONDON — A prominent UK government watchdog is weighing a formal investigation into the Home Office’s covert deployment of generative artificial intelligence to process high-stakes asylum applications. The potential probe follows mounting alarm from lawmakers, civil liberties researchers, and legal experts regarding the accuracy, transparency, and ethical implications of embedding AI models into the immigration system.

The Independent Chief Inspector of Borders and Immigration (ICIBI)—the statutory body tasked with monitoring the efficiency, consistency, and integrity of the UK’s border, immigration, and customs operations—has approached the digital rights nonprofit Open Rights Group (ORG). According to ORG campaigner Sara Alsherif, the watchdog has requested detailed findings regarding the Home Office’s AI tools as it formulates its inspection priorities for the coming year.

While the ICIBI has declined to formally confirm which subjects are under active consideration prior to publishing its annual inspection agenda, it acknowledged widespread public concern. The watchdog explicitly pointed to its own findings from last year, which concluded that the "effectiveness and reliability" of government-deployed AI tools have been remarkably "varied."

The unfolding controversy exposes deep-seated tensions between the government’s urgent drive to clear colossal bureaucratic backlogs and the fundamental human rights of vulnerable asylum seekers. Critics argue that utilizing underlying technology prone to hallucinations, cultural biases, and factual errors in matters of life and death crosses an ethical red line.

Main Facts

At the center of the controversy are two proprietary systems implemented by the Home Office: the Asylum Case Summarisation (ACS) tool and the Asylum Policy Search (APS) tool.

• The Technology: Both tools are built upon OpenAI’s 2023 GPT-4 large language model architecture and Microsoft Foundry.
• The Function: Caseworkers use the ACS to condense lengthy, complex transcripts of interviews where asylum seekers recount harrowing details of persecution and their fears of repatriation. The APS functions as a specialized search assistant, parsing and summarizing country-of-origin policy information to help caseworkers determine whether an applicant faces genuine risk of harm if returned to their homeland.
• The Stakes: According to government estimates, utilizing these tools in tandem saves caseworkers approximately one hour per case, a significant efficiency gain as the Labour administration struggles to manage tens of thousands of annual applications.
• The Secrecy: Despite the magnitude of these decisions, the Home Office has omitted both tools from the UK government’s official Algorithmic Transparency Recording Centre, a public repository mandated by the national "AI Playbook" to ensure accountability in algorithmic decision-making. Furthermore, asylum seekers are rarely, if ever, informed that AI tools have played a role in evaluating their claims.

Chronology of Events

The integration of generative AI into the UK’s immigration machinery and the subsequent pushback from civil society followed a rapid, highly opaque trajectory:

• Late 2024: The Home Office quietly launches secret pilots for the ACS and APS tools. The rollout coincides with intense political pressure on the Labour government to dismantle a staggering backlog of approximately 100,000 pending asylum cases.
• Early 2025 – Mid-2026: ORG initiates a rigorous, six-month investigation into the Home Office’s deployment of generative AI. The probe involves submitting 11 parliamentary questions, dispatching two ministerial letters, and filing 10 public records requests. The Home Office systematically withholds granular operational details.
• Mid-2026: Government statistics reveal that while the total number of individuals waiting for an initial decision has been halved through June, the volume of legal appeals and challenges against asylum rejections has ballooned exponentially.
• July 2026: A UK judge sharply rebukes the Home Office, accusing the department of relying on an "AI-hallucinated" document to reject the asylum claim of a Moroccan woman and her child on the grounds that her home country was supposedly safe.
• Late 2026: ORG and the Public Law Project (PLP) submit their dossier of findings to the Information Commissioner’s Office (ICO). The ICO launches an active investigation into the Home Office’s refusal to disclose staff training protocols and operational guidelines for the AI tools. Meanwhile, the ICIBI approaches ORG to incorporate findings into its 2027 inspection planning.

Supporting Data and Evidence

The empirical findings compiled by civil rights groups paint a troubling picture of a technology rushed into deployment without adequate guardrails or accuracy verification.

Flawed Summaries and High Error Rates

While the Home Office’s initial pilot evaluations claimed a modest 9 percent error rate for the ACS tool, ORG’s independent investigation revealed that the percentage of summaries deemed "not fit for use"—frequently plagued by severe factual errors, omissions, and outright misrepresentations—is significantly higher.

Crucially, researchers discovered an alarming absence of quality control measures for the summaries generated by the APS policy search assistant. Because these summaries dictate whether an applicant’s stated fears align with official country-of-origin guidelines, inaccuracies directly skew the legal rationale for granting or denying protection.

Legal and Regulatory Blind Spots

Independent legal opinions commissioned by ORG from Cloisters chambers suggest that the Home Office’s failure to notify asylum seekers of AI involvement—thereby depriving them of the ability to contest automated errors—may directly breach UK data protection laws.

Moreover, the Public Law Project has identified at least 11 high-risk algorithmic systems currently operating across various government departments in complete secrecy. Caroline Selman, a senior research fellow at the PLP, warns that the scale of hidden automation is likely far broader than civil society realizes.

"Other tools are being used but not being publicized at all," Selman notes. "And you can’t even begin to ask questions about things if you don’t know that they’re being used."

Official Responses

Faced with mounting scrutiny, representatives for the UK government, the tech sector, and advocacy organizations have staked out sharply contrasting positions.

The Home Office Defense

The Home Office maintains that human agency remains the absolute core of the asylum adjudication process. In an email statement to reporters, a spokesperson emphasized:

"Asylum decisions are made by trained caseworkers, with AI tools used only to assist in navigating interview records and policy information. They cannot be used in isolation and do not make decisions on claims. The use of Artificial Intelligence is just one of the tools at our disposal to protect our borders, and does not replace or automate the role of human decision-makers."

The department further noted that it regularly monitors tool performance, adheres to established quality-assurance protocols, and remains committed to publishing updated evaluations on their rollout.

Technological and Legislative Realities

OpenAI, whose GPT-4 model underpins the Home Office tools, declined to comment directly on specific government deployments. However, a company spokesperson emphasized ongoing efforts to improve model accuracy and noted that OpenAI is scheduled to officially retire the GPT-4 series models in January, transitioning users to newer architectures.

Meanwhile, critics argue that institutional reassurances ring hollow in the face of systemic opacity. Labour backbench MP Bell Ribeiro-Addy voiced profound skepticism regarding the appropriateness of automation in humanitarian contexts:

"We’re dealing with people’s lives. We’re talking about people that have faced some of the worst circumstances, and we’re meant to be making decisions based on facts and on understanding harm. I just don’t think it’s appropriate that machines are doing that."

Implications

The potential ICIBI probe and the ongoing ICO investigation carry profound implications not only for UK immigration policy, but for the broader governance of artificial intelligence within the public sector.

1. Erosion of Trust and Due Process: When asylum seekers are evaluated using summaries generated by opaque algorithms—particularly models documented to harbor cultural biases or prone to text-based hallucinations—the fundamental right to a fair hearing is severely compromised. If an applicant cannot inspect or contest a corrupted summary of their own trauma, the appeals process loses its structural integrity.
2. Accountability and the Rule of Law: The UK’s voluntary frameworks, such as the AI Playbook and the Algorithmic Transparency Recording Centre, rely heavily on departmental compliance. The Home Office’s systemic refusal to list its tools or transparently disclose training protocols demonstrates the limitations of non-binding ethical guidelines. Without statutory enforcement, government agencies can easily sidestep public oversight under the guise of operational security or bureaucratic efficiency.
3. A Litmus Test for the Watchdogs: If the ICIBI officially launches an in-depth inspection into the Home Office’s AI systems, it will set a critical precedent for how technological integration across sensitive state functions is audited. As generative AI models evolve and public sector agencies race to clear backlogs, the outcome of this clash between digital rights advocates and state bureaucracy will help define the boundaries of algorithmic governance in the United Kingdom for years to come.