The Hong Kong Police will introduce AI facial recognition to the Smart View CCTV system this month, according to The Standard. The stated aims are to detect crime and to find missing people more accurately and quickly, especially high-risk missing elderly residents with dementia. The network is targeted to reach 60,000 cameras by 2028. For anyone living in or visiting the city, Hong Kong facial recognition CCTV raises a basic question: how long will anonymity in public space last?

This post sticks to what has been reported. The source article is brief, so several important details, including technical specifications, watchlist rules and retention periods, were not included in the reporting available to us.

What the Smart View rollout changes

CCTV cameras have long recorded footage that humans review after the fact. Adding AI facial recognition changes the function of the system. Instead of a person watching video or searching it later, software can compare faces captured by cameras against a reference database, potentially in near real time.

The police have framed the upgrade around two goals: detecting crime and locating missing people. The second goal, finding high-risk missing elderly people with dementia, is the one with the most obvious public appeal. It is also worth noting that the same technical capability can serve either purpose. A system able to spot a missing person in a crowd can, in principle, spot anyone else whose face is on a list.

The scale matters too. A target of 60,000 cameras by 2028 points to dense coverage across a city. The more cameras that run matching software, the more routine movements, such as commuting, shopping or attending an event, can become searchable data points.

How the PDPO applies to police facial recognition

Hong Kong's main privacy law is the Personal Data (Privacy) Ordinance (PDPO). Its core principles cover collecting data for a lawful and relevant purpose, keeping it no longer than necessary, using it only for the purpose it was collected for, and keeping it secure. A face image that identifies a person is personal data, so these principles are the natural starting point.

The difficulty is that the PDPO was not written with live biometric matching in mind. The law is built around organizations collecting and handling records, not around a camera network that continuously scans faces of people who are not suspected of anything. Questions that a modern framework might answer explicitly remain open in public reporting so far:

  • Is there a requirement for prior independent approval before cameras are used for matching?
  • Who decides which faces go on a watchlist, and on what threshold of suspicion?
  • Are there public accuracy standards, and how are false matches handled?
  • Is there an exemption for law enforcement that limits how far the PDPO's protections reach in practice?

We cannot say how these questions are answered, because the source article does not address them. Readers should watch for official guidance from the police and the privacy regulator on how the ordinance will be applied.

Where face data goes: storage, sharing and retention questions

The most consequential privacy issues usually sit behind the camera, not in front of it. A facial recognition system creates several kinds of data: raw video, extracted face templates, match results and logs of who searched what. Each can have a different retention period and access rule.

Three questions are worth tracking as the rollout proceeds:

  1. Retention. Are faces of people who produce no match deleted immediately, or stored for later searches?
  2. Sharing. Can other agencies, or private operators of linked cameras, access matching results?
  3. Oversight. Are there audit logs and independent review of how searches are run?

Other jurisdictions have begun grappling with exactly these issues. In New Jersey, lawmakers are weighing safeguards on police use of AI and facial recognition, and the sponsors of the state's bill are seeking facts first about how the technology is actually used. That fact-finding approach is a useful template for any system that expands quickly.

What residents can realistically do to limit exposure

It is important to be honest here: individuals have limited control over a public camera network. You cannot opt out of being seen on a street. Still, a few practical steps can reduce the data trail you create elsewhere:

  • Know your rights under the PDPO. You can generally ask organizations what personal data they hold about you. Understanding this process is useful even if its reach over police systems is limited.
  • Reduce voluntary face data. Think twice before enrolling your face in non-essential apps, shop loyalty schemes or building access systems that store templates.
  • Review social media photo settings. Public photos tagged with your name can be matched with other data sources.
  • Support transparency. Responses to public consultations and questions to legislators are one of the few direct ways residents can influence rules.

A VPN is worth mentioning only to dispel a common misconception: it encrypts your internet traffic and hides your IP address, but it does nothing to hide your face from a street camera. It is a tool for online privacy, not a defense against physical biometric surveillance.

What This Means For You

If you live in Hong Kong, the practical effect is that public movement may increasingly be matched against reference databases, with the safeguards still unclear. If you travel there, assume the same. The stated purposes, such as finding missing elderly people, are legitimate and could deliver real benefits. The open issue is whether rules on retention, access and oversight are strong enough to keep a tool built for those purposes from drifting into broad everyday tracking.

Disputes elsewhere show how this debate can unfold. In India, a petition asks the Delhi High Court to declare mass surveillance of protesters unconstitutional, and Delhi Police told the Supreme Court that facial recognition at protests served a legitimate state interest. These cases are different from Hong Kong's, but they show the kinds of arguments that courts weigh.

Key takeaways

  • Hong Kong facial recognition CCTV begins on the Smart View system this month, with 60,000 cameras targeted by 2028.
  • The police cite crime detection and finding missing people, especially elderly people with dementia.
  • The PDPO sets general data protection principles, but it was not designed for live biometric matching, and key details on retention, sharing and oversight are not yet clear from reporting.
  • Individuals cannot opt out of public cameras, but can limit the face data they hand over elsewhere.
  • A VPN does not hide your face from street cameras.

To see how safeguards might look in practice, follow how courts and lawmakers elsewhere are constraining police facial recognition. The New Jersey safeguards bill and the legal challenges in Delhi over AI surveillance of protesters are good comparison cases as Hong Kong's rollout unfolds.