The New South Wales Police Force has formally deactivated the facial recognition component of its PhotoTrac Suspect Identification system, ending more than a decade of use. The decision, implemented earlier this year, came amid mounting concerns about the technology’s accuracy and significant demographic bias in its performance across different ethnic groups—though NSW Police did not cite such concerns as the reason for the system’s discontinuation.
The facial recognition software, developed by German biometrics firm Cognitec, had been in use since 2011 to compare surveillance footage against police mugshot databases. However, the software was never updated, and NSW Police continued to rely on its original legacy algorithms—despite several upgrades released by Cognitec over the years. Officials acknowledged that these older algorithms were more prone to errors, especially in identifying individuals from certain demographic groups.
The tipping point came with the release of a U.S. National Institute of Standards and Technology (NIST) Face Recognition Technology Evaluation report, which revealed stark disparities in performance. According to the report, Cognitec’s algorithm misidentified West Africans nearly seven times more often than Europeans. The false match rate for West African females aged 65 to 99 was particularly high, at 0.538 percent—the worst result across all tested demographics.
During a NSW Parliament Budget Estimates inquiry, Police and Counter-terrorism Minister Yasmin Catley confirmed that NSW Police had not purchased any of Cognitec’s post-2011 upgrades.
Despite its shortcomings, NSW Police made extensive use of the system in recent years. The technology was used to generate investigative leads in more than 560 cases during the 2023–2024 financial year and over 370 times since July 2024. Deputy Commissioner David Hudson emphasized that facial recognition was never relied upon as standalone evidence in criminal cases, but rather served as a supplementary tool during investigations.
Since 2018, the department has gradually transitioned to newer tools. These include an open-source facial recognition platform known as FaceNet, internally branded as NYX, as well as Microsoft’s AI-powered video analytics platform, Insights. While NYX continues to offer facial recognition capabilities, Insights focuses primarily on object and vehicle recognition, aligning with a broader shift away from biometric identification in sensitive applications.
The shift away from Cognitec’s system reflects growing awareness of the ethical and legal challenges posed by facial recognition in law enforcement. Globally, scrutiny around biometric surveillance has intensified, with numerous jurisdictions moving to restrict or ban such technologies. The NSW decision aligns with this cautious approach, emphasizing accountability and public trust in policing technologies.
Microsoft, whose cloud infrastructure supports NSW Police’s broader digital transformation, has also publicly committed to responsible AI practices. Since 2020, the company has declined to sell facial recognition tools to U.S. police departments, citing ethical concerns and the need for robust regulation. The AI and machine learning tools provided through Microsoft Azure are instead being used to accelerate investigations through non-biometric means, such as vehicle tracking and behavioral analysis.
While NSW authorities have not detailed the precise technical steps involved in deactivating the Cognitec system, the move is expected to have minimal impact on ongoing investigations due to the availability of alternative technologies. Still, civil liberties organizations and community advocates have called for greater transparency around surveillance tools, their use, and their eventual retirement. The legal framework governing these tools in New South Wales continues to evolve, and further oversight is anticipated as part of broader digital ethics reforms in policing.
The NSW Police Force’s decision stands in contrast to trends in other regions, where police agencies continue to expand facial recognition deployments.
Source: InnovationAus
—
April 16, 2025 – by Cass Kennedy
