180 incidents on record · 2026 Headlights Incident reports by Ellie Harris · Melbourne
10 new this week Library last updated 30 August 2026
← The incident library
HD-INC-153
Government · United Kingdom · 2020 · Algorithmic discrimination

For five years the UK Home Office graded visa applicants red, amber or green using data that included nationality, and it suspended the tool in response to a judicial review

By Ellie Harris · Filed Streaming Tool introduced 2015

Alleged: Home Office (United Kingdom) developed or deployed the AI system implicated in this incident. Details are drawn from public reports; parties are presumed innocent of any wrongdoing not established by an official finding.

For five years the UK Home Office graded visa applicants red, amber or green using data that included nationality, and it suspended the tool in response to a judicial review

What happened

It was reported that the UK Home Office introduced the automated Streaming Tool in 2015. The Independent Chief Inspector of Borders and Immigration recorded that it used Global Visa Risk Status data, including nationality, to assign visa applications a red, amber or green risk level that determined the scrutiny an application received. The inspection also recorded a ministerial authorisation allowing additional scrutiny by nationality. Applicants were not told their grades. When campaigners sought the country classifications they received a list with the country names redacted. TechCrunch reported that the Home Office had acknowledged a secret list of suspect nationalities but refused to provide meaningful information about the algorithm.

It was reported that the Joint Council for the Welfare of Immigrants and Foxglove brought judicial review proceedings alleging unlawful race discrimination under the Equality Act 2010. They alleged a feedback loop in which a higher risk grade for a nationality produced more refusals, which then raised the risk associated with that nationality. The allegation was never tested because the case did not reach a hearing, and no court determined whether the feedback loop existed or the tool discriminated unlawfully. The Home Office agreed to redesign the tool without accepting the allegations. Internal guidance suspended it on 6 August 2020 and was published six days later. The inspection report recorded that the department withdrew the tool across all entry clearance operations in response to the proceedings. Its replacement, the Complexity Application Routing Solution for Visits, does not use machine learning and does not decide applications, according to the Home Office’s algorithmic transparency record. Decision-makers assess each visitor visa application individually under the Immigration Rules and Visitor Policy Guidance.

What an auditable version would have shown

The challenge centred on whether grading by nationality produced different patterns of scrutiny and refusal among the nationalities the system sorted. No answer entered the public record, although the question was measurable. A system operating for five years could retain the distribution of applications across red, amber and green, the refusal rate for each grade, differences by nationality and changes over time. It could also preserve, for each application, the grade assigned, the factors behind it and the caseworker’s response. An applicant whose case received more scrutiny could then understand and contest the basis. Instead campaigners tried to reconstruct the system from outside and received country classifications with the countries removed.

Where the gap was

It was reported that the Streaming Tool used nationality to help determine how much scrutiny an application received. Applicants did not see their grades and the Home Office withheld the graded country list. A MetricRecord converts individual decisions into continuing population-level figures, allowing refusal rates by nationality and grade to be monitored routinely and disparities to become internal governance signals rather than matters first raised in court. A ConductRecord preserves what the system did in each case and how the human decision-maker responded, allowing an applicant to understand and contest the grade. The records answer different questions, one about the system’s effects across a population and the other about its effect on one person. The tool operated from 2015 until litigation prompted its withdrawal in 2020.

What governance should have looked like

It was reported that the UK Home Office did not accept the challengers’ allegations, and it was reported that it never had to answer them at a hearing because it withdrew the tool first. Where the state uses an automated grade to determine the scrutiny a person receives, governance needs population-level measures showing how grades and outcomes differ among the groups being sorted, with review triggered by a material disparity, and an individual record of why a grade was assigned, how it shaped the process and who remained responsible for the decision. Legitimate limits on public disclosure do not remove the need for internal scrutiny or a route to challenge. The replacement routing solution does not use machine learning. Five years of grading ended with serious but untested allegations, a withdrawn system and no published measurement showing what the grading had done.

Failure Pattern: an automated risk grade that took nationality as an input decided how much scrutiny a person’s application received, while the person could not see their grade and the criteria were withheld.

Governance Principle: where an automated grade determines the scrutiny a person receives from the state, its performance across the groups it sorts should be measured and published, and the basis of each grade should be recorded and open to challenge.

The reference implementation of MetricRecord and ConductRecord is open source. It lives at github.com/saffronandindia/headlights-oss, Apache 2.0 licensed and free to install. The repository is public now.

Sources

The mailing list

Fresh incident reports every week. One email to match.

We add new incidents to the library regularly, and send a single short email each week with what's new. The library stays free and open; this is just how you keep up with it.

No tracking. Unsubscribe in one click.

The record

An auditable system would have produced a signed, tamper-evident record the moment this happened: what the system did, the version that did it, the basis it acted on, and the action taken, and Home Office (United Kingdom) could have produced it on demand.

This is the record the system as deployed did not produce in a signed, auditable form.

What this teaches
Capture what happened when it happens
What the system did, the version that did it, the basis it acted on, and the action taken, recorded at the moment, not reconstructed after.
Sign it, so no one has to trust the record-keeper
A tamper-evident entry. Edit it later and the signature breaks. The record does not ask for the benefit of the doubt.
Make it verifiable by anyone
A court, a regulator, a customer's lawyer can check the record themselves, without taking the company, or us, at our word.

Headlights summarises publicly reported AI incidents. All summaries are independently written, attributed to their original sources, and intended for research and educational purposes. Allegations are identified as such until established through official findings.

This report is based on the sources listed above and reflects information available at the time of review; later developments may not be captured. Where a person is described as charged with or alleged to have done something, that allegation is unproven unless a conviction or a court or regulatory finding is stated. Headlights publishes journalism and commentary, not legal advice.

Want to write back?

Direct to my inbox.

ellie@useheadlights.com →