A six-month trial of live facial recognition (LFR) technology conducted at railway stations in London, which incurred costs exceeding £320,000 and utilized nearly 100 hours of police manpower, concluded with only a single incorrect match against a suspect watchlist and no arrests made.
From February to July of this year, over half a million facial scans were performed at some of London’s busiest transport hubs as part of a British Transport Police (BTP) initiative aimed at identifying offenders and individuals violating court orders.
A freedom of information request obtained by Liberty Investigates and shared with the Guardian revealed that the expenses associated with equipment rental and police staffing for the 18 deployments totaled £320,786. This investment resulted in just one alert from the watchlist, which ultimately proved to be a false positive.
Recently, the BTP announced an extension of the trial for an additional four months, broadening its scope to include London Underground stations. Transport for London endorsed this extension, asserting that the LFR technology is designed to target and identify individuals on police watchlists at strategically selected stations to maximize effectiveness. The initiative aims to specifically address violence against women and girls, whose travel experiences are often influenced by incidents of sexual harassment and offenses.
Fraser Sampson, former UK biometrics and surveillance camera commissioner, emphasized the need for the police to demonstrate that the application of this technology is justified. He noted that numerous factors need to align, including the choice of locations, time of day, composition of the watchlist, and the likelihood of individuals being present.
Sampson, now a non-executive director at Facewatch, a company specializing in facial recognition systems for retail, remarked that while the technology is effective, its application in retail environments to deter theft and protect staff differs significantly from its use in public transport settings. He stated, “Success in a shop means preventing unwanted individuals from entering, whereas success in law enforcement means actively apprehending suspects. In that regard, the trial does not seem to have produced significant results.”
He further explained that the efficacy of the LFR cameras is contingent on the watchlists; individuals not listed will not be recognized or retained by the system, unlike traditional CCTV cameras. However, he stressed that the technology processes sensitive personal data and must be employed in an appropriate, proportional, and necessary manner. It is the responsibility of the police to demonstrate how their deployment complies with these standards, and oversight bodies must hold them accountable, as highlighted in a recent report by the Information Commissioner’s Office (ICO).
According to an analysis by Liberty Investigates, more than half of the police forces in England and Wales have implemented LFR technology in public spaces. This past summer, the Metropolitan Police and the London Mayor’s office announced plans to install fixed LFR cameras in the West End, with Mayor Sadiq Khan confirming the technology’s use on the newly pedestrianized Oxford Street.
A recent report from the parliamentary joint committee on human rights urged the government to establish regulations addressing the potential risks of artificial intelligence to human rights, citing the deployment of live facial recognition technology as a particularly evident concern.
Labour MP Bell Ribeiro-Addy expressed her concerns, stating, “Live facial recognition technologies represent a significant intrusion into our privacy. The government should halt their deployment until stronger protections and a comprehensive legal framework are established.”
A spokesperson for the BTP stated that the purpose of the LFR pilot was to assess how effectively the technology could assist in identifying wanted offenders and individuals posing risks to passengers and staff. They noted that while several arrests were made during the trial for offenses such as assault, theft, and public disorder, these arrests were not a direct result of LFR alerts and thus were not included in the performance data for the technology.
Throughout the trial, the BTP indicated that it has been refining its deployment methods within the railway environment, which includes adjustments to deployment locations, operational procedures, equipment, and watchlist development. They underscored the importance of deliberate and controlled growth of the watchlist, ensuring that data quality, governance, and safeguarding measures remained robust from the outset.
