Open Rights Group says AI is embedded across the Home Office with little public scrutiny

Campaigners say AI now runs through many Home Office tasks

The Open Rights Group (ORG) has published a report, titled Automated Discrimination, arguing that AI is embedded across the Home Office in near-total secrecy. According to the group, the technology is used to flag possible sham marriages, estimate the age of asylum seekers and check the contents of phones belonging to new arrivals. It also supports border surveillance through live facial recognition and AI-powered drones and satellites, and helps prioritise and assess cases through a tool called Identify and Prioritise Immigration Cases (Ipic). ORG says its findings were assembled from limited Home Office publications, freedom of information requests and questions from MPs. The report argues that these systems are neither neutral nor accurate and can reproduce bias and discrimination.

Cross-party backing comes from Labour MP Bell Ribeiro-Addy, the SNP's Pete Wishart and the Green party's Siân Berry. Ribeiro-Addy has called for the Home Office to publish its equality and data protection assessments and to allow independent validation of how well the tools perform.

“The Home Office should not be using AI in decisions this fundamental to people's lives until it can show the tools are accurate”

The campaigners want the government to regulate AI, ban uses that risk human rights violations, publish a public register of government AI tools and tighten freedom of information laws.

Asylum tools and accuracy concerns

The Home Office relies on two systems, the Asylum Case Summarisation (ACS) and Asylum Policy Search (APS) tools, which draw on OpenAI's 2023 GPT-4 model and Microsoft Foundry. Documents obtained by ORG show that the tools summarise transcripts of interviews with asylum applicants and background material on applicants' countries of origin. The immigration authority began piloting them in late 2024, when the government was working through a backlog of around 100,000 asylum cases. The department has since halved the number of people waiting for an initial decision in the year through June, while the number of appeals has risen sharply. The Independent has reported that internal spot-checks found the summarisation tool failing basic accuracy tests.

“Asylum decisions are made by trained caseworkers, with AI tools used only to assist in navigating interview records and policy information.”

The department says it monitors the tools alongside its quality-assurance processes and has committed to publishing updated evaluations as they roll out.

Watchdog review and the secrecy question

The Independent Chief Inspector of Borders and Immigration (ICIBI) has asked ORG for more information on the tools while it drafts its inspection priorities for next year. The inspectorate would not confirm which topics are under active consideration, but said it is aware of concerns about AI use and pointed to its earlier findings that the tool's effectiveness and reliability were varied. ORG's six-month investigation drew on 11 parliamentary questions, two ministerial letters and 10 public records requests, because the Home Office has withheld details about the tools. Sara Alsherif, an ORG programme manager, has said that transparency, accountability and strict regulation are urgently needed to limit how AI is used and to prevent its wider rollout.

Parliamentary questions11

Ministerial letters2

Public records requests10

Parliamentary questions

11

Ministerial letters

2

Public records requests

10

Data from the article text

Datacentre policy and the revolving door

Spotlight on Corruption found that 38 ministers, senior civil servants and special advisers moved into positions at AI companies between July 2024 and September 2026, including roles at OpenAI and Anthropic. Its investigation filed 40 freedom of information requests to 18 government agencies over 12 months and covers the AI growth zones policy, a central part of former prime minister Keir Starmer's AI and economic agenda. Nine of the departing officials came from the Department for Science, Innovation and Technology, among them a former minister for AI and intellectual property and a former AI adviser to the prime minister. Ministers, permanent secretaries and directors general involved in the policy met 137 businesses, some of them several times, while senior officials met two community groups and no environmental organisations.

“The government totally failed at the most senior level to speak to environmental NGOs or community groups when developing its flagship policy on AI datacentres.”

Business meetings137

Community group meetings2

Environmental group meetings0

Business meetings

137 meetings

Community group meetings

2 meetings

Environmental group meetings

0 meetings

Data from the article text

Wider AI harms in the same coverage

UK anti-slavery commissioner Eleanor Lyons has warned that AI makes it easier for criminals to build fake online profiles and groom children into carrying money or drugs. She said victims are often out of reach of authorities. New data cited in the coverage shows 7,045 children were referred as potential victims of trafficking and modern slavery in 2025, up 17% on the previous year, with more than 60% of them British nationals. The Internet Watch Foundation says sextortion of under-18s is also rising. Separately, Microsoft UK's chief security adviser Dr Jon Davies said AI-driven threats are real, and the company's Digital Defense Report found phishing attacks rose globally from 7% to 23% of incident-response intrusions year-over-year. Google, meanwhile, has accused the Competition and Markets Authority of an "unlawful" attempt to make it promote rival AI services.