[2024] UKUT 71 (AAC)

In response to PLP’s information request, HO sent, amongst other things, a 15 page document (the “EIA”) headed “Borders Immigration Citizenship Systems/Equality Impact Assessment”. The EIA set out the “equalities considerations” for introduction of an automated system to be used to triage referrals under the sham marriage referral and investigation scheme. It stated that sham marriages allow individuals to gain an “immigration advantage” to which they are not entitled; it said that sham marriages were linked in “many” cases to wider organised crime; which is why, it said, HO focus on “disrupting facilitators” as well as prosecuting individuals involved in sham marriages.

The EIA stated that information on referral under the sham marriage referral and investigation scheme is assessed against the “risk factors” in the automated triage system, being “eight specific criteria”, to determine the potential risk of the couple engaging in a sham marriage.

On page 6, there was a redacted section, following the words: “The triage process uses the following eight criteria.”

Following this, the EIA stated as follows:

“An extensive model training exercise was undertaken in 2018-19 to select the most appropriate features as defined above.

It is important to note that the model works by combining the predictive power of all the features. For instance, a larger value of one feature doesn’t necessarily mean there is a stronger correlation of a pass or fail of the triage.

The most powerful features identified related to the interactions between the couple, reflecting the fact that the model is predicting the likelihood of sham marriages. In the case of the Section 24 features, these are generated from the Section 24 form which is populated by the registrar observing the behaviour of the couple. Registrars meet large numbers of people giving notice to marry and are able to ascertain unusual behaviour. Another important feature is shared travel events, whereby if a couple have travelled on many flights together there is a suggestion of a stronger relationship between the parties.

Other considerations included the independence of the features and the avoidance of protected characteristics. Age is a protected characteristic, but it is important to consider the at the model does not use an individual’s age, but instead uses the age difference in days between the couples ages.”

HO cited section 31(1)(a) and 40(2) for the redactions in the information it provided.

PLP, in response, sought “internal review” of the redactions made by HO, as well as of HO’s “failure to disclose its full analysis of the impact of the triage model on different nationalities”. This was on the basis that the EIA stated on page 9 that “a review of the nationalities involved has been conducted” and page 10 included a graph of the impacts on different nationalities (which, PLP said, was “incompletely labelled”). PLP said that it wanted a “review” of HO’s failure to disclose “all of its assessments of the equality impacts” of the triage model, “as originally requested”.

HO responded (22 June 2021), saying that s31(1)(a) was engaged to withhold the criteria that the triage model uses; and that HO did not hold any further information on the point about the impact of the triage model on different nationalities.

Later in the process (after PLP had made a “complaint” to IC under s50), HO provided a version that disclosed two of the triage process criteria listed on page 6 of the EIA: “number of shared travel events”, and “age difference between the couple”.

HO said this to IC (email of 22 September 2022) about its disclosure of the “age difference” criterion:

“I considered redacting this section. This is the only triage criteria that touches on a protected characteristic. There is the potential for there to be some indirect discrimination based on age. I felt it was in the public interest to understand how we justified this potential indirect discrimination. The risk of disclosing one criteria was outweighed by the public interest in understanding this impact. The other criteria do not use protected characteristics and therefore the risk of disclosing them outweighed the public interest.”