What happened
Hong Kong police announced that they will activate an AI‑powered facial‑recognition capability across their SmartView surveillance network on Monday. The system, launched as part of the SmartView project in 2024, will allow officers to search for individuals in recorded footage and receive real‑time alerts when a match appears on live cameras. Searches require approval from an inspector‑rank officer and must be justified for national security, crime prevention, or public safety. Video footage will be retained for up to 31 days, while facial‑recognition metadata – timestamps and locations of matches – may be stored for up to 180 days. The rollout follows earlier pilot uses, including a private residential estate that announced facial‑recognition access control, prompting a privacy‑watchdog compliance check.
During a police passing‑out ceremony, Commissioner of Police Joe Chow Yat‑ming confirmed that the AI facial‑recognition system would go live on Monday. The technology is part of the SmartView project, which was launched in 2024 with a goal of installing 60,000 cameras across Hong Kong by 2028. Chow said the system can identify individuals even when they wear masks or attempt other evasion tactics, positioning it as a deterrent for potential offenders.
The system offers two primary functions. First, officers can input a suspect’s facial features to search archived footage, retrieving instances where the person appeared on specific cameras within a chosen timeframe. Second, a real‑time mode can alert officers when a match occurs live, a capability the force highlighted for locating missing persons, such as elderly residents who cannot be found in recorded video.
Operational safeguards require that any facial‑recognition search be authorised by an officer of inspector rank or higher, and the request must be tied to safeguarding national security, preventing or detecting crime, or ensuring public safety. Video recordings will be retained for a maximum of 31 days, while the system’s metadata—records of when and where a person was identified—may be stored for up to 180 days because it occupies less storage space.
The technology is already being trialled beyond the police force. The management office of Laguna City, a private residential estate in Lam Tin, announced plans to use facial‑recognition for building access control. Residents expressed data‑privacy concerns, prompting the city’s privacy watchdog to launch a compliance review of the estate’s system.
Why it matters
The deployment marks one of the most extensive uses of AI‑driven facial‑recognition by a municipal police force, expanding a network that will eventually encompass 60,000 cameras citywide. By enabling both retrospective searches and real‑time alerts, the technology could significantly increase police investigative speed and deterrence, especially in cases involving masked suspects or missing persons. However, the move also raises substantial privacy and civil‑rights concerns. Retaining match data for six months, even if video is deleted after a month, creates a long‑term record of individuals’ movements that could be misused or exposed. The requirement for senior‑officer approval offers a procedural safeguard, but the criteria—national security, crime prevention, public safety—are broad and may be interpreted expansively. Internationally, similar systems have faced criticism over racial bias, false positives, and lack of transparent oversight. Hong Kong’s adoption therefore adds to the global debate on balancing law‑enforcement benefits against potential infringements on privacy and non‑discrimination principles.
The scale of Hong Kong’s rollout makes it a for AI‑enabled public‑surveillance projects. With a target of 60,000 cameras, the city will have one of the densest AI‑augmented surveillance networks in the world, potentially reshaping policing tactics and public expectations of safety.
Retention of facial‑recognition metadata for six months creates a longitudinal record of individuals’ movements that could be accessed by multiple agencies or, if compromised, by malicious actors. This raises questions about data security, purpose limitation, and the risk of mission creep beyond the stated law‑enforcement uses.
Internationally, facial‑recognition deployments have faced scrutiny over , especially against minority groups. Hong Kong has not yet released performance metrics or bias assessments for its system, leaving a gap in public accountability and making it difficult to evaluate the technology’s fairness.
The requirement for senior‑officer approval is intended to provide oversight, but the broad justification categories—particularly “national security”—could be invoked in a wide range of scenarios, potentially limiting the system’s use as a proportionality safeguard.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Why can ethical evaluation not be reduced to one model score?
What to watch next
Watch for how Hong Kong’s privacy watchdog evaluates the residential‑estate pilot and whether it issues broader guidance for the citywide rollout. Monitor any legal challenges or public protests that may arise, especially concerning data‑retention periods and the scope of “national security” justifications. International observers will likely compare Hong Kong’s framework to emerging regulations in the UK, US, and EU, potentially influencing future policy harmonisation. Finally, track the accuracy and bias metrics the police publish, as these will affect public trust and could adjustments to the system’s deployment.
The privacy watchdog’s compliance check of the Laguna City pilot will likely set precedents for data‑protection standards that could be applied to the broader police network.
Legal challenges may emerge if individuals allege wrongful identification or misuse of stored metadata. Courts’ handling of such cases will inform the legal boundaries of AI surveillance in Hong Kong.
International bodies, such as the UN Human Rights Council or the EU’s data‑protection regulators, may comment on Hong Kong’s approach, influencing diplomatic or trade considerations related to privacy standards.
Future disclosures from the police about system accuracy, false‑positive rates, and bias testing will be critical for assessing whether the technology can be trusted to operate without disproportionate impact on vulnerable groups.