Attributed metadata history

UK sets rules for live facial recognition in immigration enforcement

The UK has published updated guidance governing Immigration Enforcement's use of live facial recognition to locate people of immigration law enforcement interest, setting out rules for watchlists, deployment locations, human oversight and biometric matching thresholds. The Home Office policy covers operations beyond ports of entry, including transport hubs and joint deployments with police, while establishing safeguards intended to ensure that facial recognition is used lawfully and proportionately. The new guidelines instruct IE staff on “how to use overt LFR to locate relevant people lawfully and ethically” and “support IE’s and the Home Office’s legitimate immigration enforcement aims.  It provides a summary of the operational use of LFR and the governance and oversight applied. “IE staff must follow the agreed LFR policies and processes,” it says. “This helps IE use LFR lawfully and effectively, with appropriate safeguards, and helps maintain trust and confidence among the public, partners and other stakeholders.” Deployments must be contextual, proportional, time-limited To that end, the document lays out step-by-step protocols on “how the IE Live Facial Recognition (LFR) policy works from lawful purpose to accountable deployment,” in addition to a host of safety and governance considerations. Defining lawful purpose comes first. Then, it is determined who is eligible to be on an LFR watchlist. Watchlists are compiled with safeguards applied, requiring them to be linked to a defined deployment objective, location and context; time limited; and proportional, with a necessity and proportionality assessment completed. Use must be overt and subject to human decisions about engagement or enforcement. Images of people who do not generate an alert are “automatically and permanently deleted,” and deployments must be “monitored, reviewed and accountable.” Per the document, IE’s Emerging Technology Team (Strategic Services and Transformation) provides senior internal oversight for LFR. The policy states that “transparency is the default position for Immigration Enforcement’s use of LFR. Any departure from advance notice is exceptional, must be justified and recorded, and does not affect the requirement that deployments remain overt and subject to post-deployment publication and scrutiny.” The policy also makes clear that Immigration Enforcement's use of live facial recognition is not confined to ports of entry. Deployments may take place at transport hubs, along travel routes and at events, subject to assessments of necessity, proportionality and operational risk. Higher-risk deployments require additional oversight. The guidance also provides for cooperation with police forces, including joint deployments in which Immigration Enforcement can supply watchlists for police facial recognition systems or process police-supplied data through its own systems. The arrangements extend the potential operational reach of facial recognition in immigration enforcement while requiring deployments to remain subject to the policy's safeguards. Testing gets specific on false alert rate metrics The guidance is extensive, and includes key performance metrics on false alert rate (FAR) for the biometric technology powering live facial recognition: operational experience suggests that, in most scenarios, FAR should generally be 0.1 percent or lower, equivalent to fewer than one false alert per 1,000 comparisons. It also cites testing done in 2021 by South Wales Police (SWP) and the Metropolitan Police Service (MPS), to “test the accuracy and equitability of facial recognition tech in operational settings,” and specifically to a report on the NEC system used by the National Police Chiefs Council (NPCC) police forces. “After considering the report’s findings, engaging with UK police facial recognition leads, and deploying LFR with police support, IE will continue to use a threshold of 0.64 for IE-led

This record contains metadata and attribution only. It does not host, download, unpack, render, or execute malware samples.