CaraComp
CaraComp
Forensic-Grade AI Face Recognition for:
Get Started7-day refund guarantee**

Biometrics and Biometric Data: Hong Kong's 2,800 Arrests

biometrics hong kong facial recognition street cameras with a face outline and a question mark overlay
Hong Kong facial recognition cameras and a question mark show how biometrics checks can get a person wrong. Illustration: CaraComp

Quick answer

Can facial recognition flag the wrong person by mistake?

Yes. Facial recognition compares two faces and produces a match score, and look-alikes can score high enough to be flagged. A study of police use in Wales found false positives happened over ten times as often as correct identifications. A match works best as a lead that a human checks, not as proof.

Picture a stranger's face on a screen, and a computer saying it matches yours. You did nothing. You were just walking through a shopping mall. That is the worry sitting underneath the news about biometrics this week (biometrics means your face, voice and fingerprints, each a measurable physical characteristic that is uniquely you).

TL;DR

Biometrics are turning into everyday safety checks, from Hong Kong's camera network to teen age checks, and a computer's guess is not proof, so people need a clear way to challenge a wrong match.

Three stories about biometrics landed in the same week. Hong Kong is switching on AI facial recognition across a growing network of cameras. Instagram's parent company is rolling out AI that guesses whether a user is really a teenager. And Malaysia's government is studying whether people should have to prove who they are before using WhatsApp, to fight scams. Different countries, same question: what happens when the system gets you wrong?


Biometrics in Hong Kong: Face Recognition and Biometric Data on the SmartView Network

Here is the short version of the Hong Kong story. Police run a camera program called SmartView, which started in 2024. The plan is about 60,000 cameras across the city by 2028. According to Biometric Update, the system is getting real-time alerts (a ping the moment a face seems to match someone on a list), links to shopping mall cameras, and 180 days of stored footage data.

The South China Morning Post reports the police announcement put the AI activation on November 16. Mall camera links are expected by the end of 2026.

2,800+
arrests police say SmartView has contributed to since launch, with no public count of false matches
Source: Biometric Update

Police say SmartView has helped solve nearly 2,200 criminal cases. Those numbers sound impressive. But notice what is missing. Nobody has published how many times biometric systems pointed at the wrong person. A scoreboard with only the wins on it is not a scoreboard.

What Hong Kong police say about how face recognition biometrics will be used

In fairness, police say the face matching will be switched on for specific searches, such as missing persons and named suspects, not run on everyone all the time. They point to limits on computing power. Any search needs sign-off from an officer ranked inspector or above. And the privacy watchdog says organizations must follow Hong Kong's privacy laws, a point Hong Kong Free Press covered when police said the mall link-up complies with those laws.

That is a real safeguard. It is also a safeguard about who may start a search. It says nothing I can find about what happens after a match: who double-checks it, and whether the person flagged ever gets to push back. This article is part of a series, start with Minimum Act Score Colleges Vary Eu Social Media Age Is 15.


Why Do Biometrics Get People Wrong? The Pros and Cons of Face Recognition

Face recognition makes mistakes because it does not "see" a person. It compares two faces and produces a match score (a number for how alike they look). Someone then picks a cutoff. Set it loose and you catch more suspects, but you also catch more innocent look-alikes. That is called a false positive (the computer says "match" and it is wrong). Biometrics of this kind always trade one kind of error for another.

A study of real-time facial recognition by police in Wales found that false positives occurred over ten times as often as accurate identifications.

Finding on the Wales study, as summarized by the Project on Government Oversight

Read that twice. In that study, wrong flags outnumbered right ones by more than ten to one. And the errors are not spread evenly. According to a 2019 test by NIST (the US government's technology standards lab), false matches for Asian and African American faces were often 10 to 100 times more common than for white faces, depending on the program. The Project on Government Oversight also digs into those demographic gaps. I will be careful here: I am not saying Hong Kong's system performs this way. Nobody outside can say, because there is no published accuracy testing. That is exactly the problem.

How Hong Kong face recognition could turn a lead into a verdict

Here is the pattern worth watching. A match is supposed to be a lead, a reason to look closer. But leads have a way of hardening. An officer sees a screen say "match," feels pressure to act, and a guess starts behaving like proof. Nobody announces that shift. It just happens.

Cons for Ordinary People: Why Biometrics Mistakes Matter

  • Innocent people get stopped because a look-alike triggered an alert, and being stopped costs time, trust and sometimes more.
  • Errors are uneven because some groups face far higher false match rates than others.
  • Records pile up because 180 days of stored data is a long time to hold footage of people who did nothing.
  • Appeals are unclear because no public rule says how a person flagged in error can challenge it.

Trusted by Investigators Worldwide
Run Forensic-Grade Comparisons in Seconds
Detailed facial comparison reports. Results in seconds.
Get Started
7-day refund guarantee**

Teen Age Checks and WhatsApp Identity: Biometric Authentication in Daily Life

Now step away from police. The same idea is showing up in your kid's phone. Instagram is expanding AI age checks worldwide, aimed at spotting children who sign up pretending to be adults. That means age-estimation technology is guessing a teenager's age, often from how a face or account looks, which is a close cousin of biometrics. A guess is all it is. A 15-year-old who looks 19 might slip through. A 19-year-old who looks young might get locked out.

Malaysia is looking at a different version. Its government is studying whether WhatsApp users should have to verify their identity, according to The Straits Times, as a way to cut down on online scams. Parliament has also been debating social media age checks. The goal in each case is protection. The risk in each case is the same too: a wrong answer, with a real person on the receiving end.

Where biometrics show upWhat the system guessesWhat a wrong answer looks like
Hong Kong SmartView camerasWhether a face matches a suspect or missing personAn innocent person flagged as a suspect
Instagram AI age checksWhether a user is really a teenA teen treated as an adult, or an adult treated as a child
Malaysia WhatsApp ID studyWhether an account belongs to a verified real personA real person locked out, or a scammer let through

Look at that last column. Every row ends the same way: a person caught in a decision they never asked for. And in every row, the real fix is the same too. A human has to be able to review the result, and the person affected has to be able to ask for that review. Previously in this series: Age Verification Nairobi Court Rejects Typed Birthdays.

What should a person do if a system flags them or their child by mistake?

Start by asking three things in writing: what exactly was flagged, who reviewed it, and how to appeal. Keep screenshots and dates. Ask for a human to look at it. Systems built with real appeal paths will have clear answers. If nobody can tell you, that silence is information.


What Can People Do About Biometrics Errors Before They Happen?

If you have ever wondered whether a photo or a profile is really who it claims to be, that is the exact question biometrics exist to answer. The honest answer is that biometrics can reduce security risks, but a match is only as good as the checking behind it. So here is one thing you can actually watch for: does anyone say how often the tool is wrong?

A trustworthy tool comes with a documented error rate and a record of who looked at what, and when (an audit trail). A second person should confirm any match before anything serious happens. If a system or a service offers none of that, treat its answer as a hunch, not a fact. That goes for a city camera network, a school app, or a work ID check.

And the strongest point from Hong Kong's rollout? Safeguards that cover who can start a search are not the same as safeguards that protect the person the search finds. Sign-off from an inspector is a door lock. An appeal right is a smoke alarm. You want both.

Key Takeaway

Biometrics now touch policing, teen accounts and messaging apps, and in each place a face match or age guess is a clue, never proof, so a person should always be able to ask a human to double-check.

Recap: Face and age checks are becoming part of normal life, and a computer's guess can be wrong. People need a simple, real way to challenge a mistake.

So here is the line I keep coming back to. Hong Kong can tell you how many arrests its cameras helped with: more than 2,800. It cannot yet tell you how many people those cameras got wrong. Until it can, that number is only half a number. Up next: Biometrics Hong Kong Counts 2 800 Arrests Not Its Errors.

Biometrics: Frequently Asked Questions

What are biometrics in simple terms?

Biometrics are measurements of your body that are unique to you and used to recognize you, like your face, voice, fingerprints or eyes. Companies and governments use biometric information to verify someone's identity or to identify a person in a crowd. It feels convenient because you cannot forget your face the way you forget a password. But that is also the risk: you cannot change your face if your biometric data is mishandled or a system matches it wrongly.

What is Hong Kong face recognition and SmartView?

SmartView is a Hong Kong police camera program that began in 2024 and aims for about 60,000 cameras by 2028. Reports say AI face matching is being activated across the network, with real-time alerts and links to shopping mall cameras expected by the end of 2026. Police say the biometric matching will be used for specific searches, not constant scanning of everyone, and that searches need senior officer approval.

Can biometric systems really flag the wrong person?

Yes. Face recognition compares faces and produces a match score, and look-alikes can score high. A study of police use in Wales found false positives happened over ten times as often as correct identifications. A 2019 NIST test found error rates varied a lot by demographic group. This is why a biometric match should be treated as a lead that a human checks, not as proof.

Is an AI age check the same as an ID check?

No. An AI age check estimates whether someone is a teen or an adult, usually by analyzing how they look or behave, much like biometrics do. An ID check compares a person against an official document or record. The estimate is a guess and can be wrong in either direction. An ID check is usually stronger, but it asks people to hand over more personal information, which brings its own privacy worries.

How long does Hong Kong keep SmartView data?

The reporting says the SmartView system will keep data for 180 days. That is roughly six months of stored footage data. Critics argue this is longer than most investigations need and starts to look like an archive of ordinary people's movements. Police say the system complies with Hong Kong privacy law, and the privacy watchdog has said organizations must follow those laws.

What should I ask if a system wrongly flags my child?

Ask what was flagged, which tool made the call, whether a human reviewed it, and how to appeal. Request answers in writing and keep screenshots with dates. A well-run service should have a clear process and a real person who can reconsider the result. If nobody can explain how to challenge the decision, that is a warning sign about how the check was built.

Ready for forensic-grade facial comparison?

Full forensic reports with detailed similarity scoring. Results in seconds.

Run My First Search