LEDE
Three justice agencies have put face checks into community supervision: the Ministry of Justice’s Probation Service in England and Wales, the Massachusetts Probation Department and Queensland Corrective Services. All three use the face match as one extra check on top of supervision they already ran: officers, tags, appointments and assessments. Their public material describes how each confirms who is checking in. None of it says how each one stops a photo, a screen replay or a mask from passing as the person, and that is the gap a buyer should close first.
CONTEXT
Why Identity Checks Matter In Community Supervision
A remote check-in only helps an agency if it proves who checked in. Location data, a breath sample or a reported address says nothing useful when someone else gave it on the supervised person’s behalf.
Community supervision covers probation, parole, licence conditions and curfews. Supervision has always depended on face-to-face contact, and contact costs officer time. Remote check-ins move part of that contact onto a device. The cost is a new question that an office visit answered by itself: is the person in front of the camera the person on the order? Face verification answers part of it, by matching the face to an enrolled record. Liveness detection answers the other part, by checking that the face belongs to a live person at the camera and is not a printed photo, a video on a screen or a mask. ARSA’s guide to how active liveness detection works explains the challenge-response method in more depth.
CASE ONE
England And Wales: The Probation Service Pilots Remote Face Scanning
The Ministry of Justice is piloting remote video check-ins in which AI confirms the offender’s identity. The check-ins sit on top of existing licence conditions.
The Problem
Probation officers in England and Wales already manage people on licence through GPS tags and in-person appointments. The Ministry of Justice describes the remote face scanning as a way to add contact between those appointments, and it presents the pilot as more monitoring that leaves the existing conditions in place (GOV.UK, Ministry of Justice).
The Approach
The pilot adds remote video check-ins with AI identity confirmation, and the GPS tags and in-person appointments stay (GOV.UK, Ministry of Justice). UKTN reports funding of £8m and a pilot running in a handful of probation regions in England (UKTN).
The Result
The pilot was announced on 10 September 2025, and neither source publishes results. Match rates, false rejection rates, the anti-spoofing method and where the face data is processed and stored are not public in either source.
CASE TWO
Massachusetts: Probation Pairs A Breathalyzer With A Face Camera
The Massachusetts Probation Department used a handheld breathalyzer that photographs the person at each test and matches the photo to their enrolled face.
The Problem
A remote sobriety test only means something if the person on probation is the one blowing into the device. With a plain breathalyzer, a sober friend could give the sample.
The Approach
WHDH 7 Investigates reported that the device takes a high-resolution photo during each test, matches it against the enrolled facial record to confirm identity, and records the GPS location of the test, which it reports to the Massachusetts Electronic Monitoring Center (WHDH 7 Investigates). Route Fifty identifies the device as the SCRAM facial-recognition breathalyzer (Route Fifty).
The Result
About 700 people on probation in Massachusetts were being monitored with the device as of the 2017 report (Route Fifty). Neither source gives match accuracy, how often a test failed identity, or whether the camera checks that it is looking at a live face and not a photograph.
CASE THREE
Queensland: QCS Connect Puts Biometrics In A Kiosk And An App
Queensland Corrective Services built QCS Connect, which uses fingerprint and facial biometrics in an in-office kiosk and in a mobile app so that eligible people can self-report and answer curfew checks.
The Problem
Reporting and curfew checks take officer time, and eligible people under supervision can do part of that contact themselves if the agency can confirm who is reporting.
The Approach
QCS Connect runs on fingerprint and facial biometrics in two places: a kiosk in the office and an app on a phone. A Community Corrections officer assesses each supervised person’s access on its own, based on that person’s needs and circumstances (Queensland Corrective Services).
The Result
The service is listed as available. The page gives no launch date and no usage or performance figures (Queensland Corrective Services).
| Agency | Where The Face Check Happens | What It Adds To | Published Figures |
|---|---|---|---|
| Probation Service, England And Wales | Remote video check-in | GPS tags, in-person appointments | £8m funding, a handful of regions (UKTN) |
| Massachusetts Probation Department | Handheld breathalyzer | Breath test, GPS location | About 700 people monitored (Route Fifty) |
| Queensland Corrective Services | Office kiosk and mobile app | Self-reporting, curfew checks | None published |
READING THE CASES
What The Three Deployments Have In Common
All three agencies use the face check to confirm one event at a time: a check-in, a breath test or a curfew answer. None of them uses it in place of supervision.
Three lessons carry over to any buyer.
Attach The Face To The Event
Massachusetts takes the photo during the breath test, so the identity and the sample come from the same moment. An app that checks identity at login and lets the person hand over the phone afterwards proves much less. Bind the verification to the action you need proof of.
Keep The Human Decision
Queensland leaves eligibility to an officer, one person at a time, and England and Wales keep the existing licence conditions. A failed face match should go to an officer as a prompt, and the agency should decide in advance what happens after one.
Ask About Spoofing And Error Rates
None of the five sources says how the system tells a live face from a photo or a replayed video, and none gives a false rejection rate. In supervision a false rejection can look like a missed check-in, so ask for that figure before you buy.
What is still unknown matters just as much. No source here says where the face images are processed or who holds the templates. That question has legal weight in every market this article addresses. UK GDPR treats biometric data used to identify a person as special category data. Queensland agencies work under the state’s Information Privacy Act. US criminal justice agencies that connect to FBI systems answer to the CJIS Security Policy. A vendor cloud adds a party to each of those obligations.
ARSA’S ANGLE
How ARSA Approaches Remote Identity Verification
The ARSA Face Recognition and Liveness SDK runs face matching and liveness checks inside the agency’s own infrastructure. ARSA did not build or supply any of the three deployments above. What follows describes the product as its product page does.
The SDK ships as one container. It holds the face database, 1:1 verification, 1:N identification, passive anti-spoofing on a single image and active liveness over video. In active liveness the person performs a randomised sequence of head movements, and the system checks both the order of the movements and the anti-spoof result. One call, submit_and_recognize, returns the liveness verdict and the identity together, and it skips recognition entirely when liveness fails, so a spoofed check-in is never matched to anyone. The licensed product makes no outbound call at inference time and sends no telemetry. Templates are never exposed over the API. Request logs record the method, status, latency, masked key and client IP, and never the request or response bodies.
The agency or its integrator builds the check-in app. The SDK supplies the identity and liveness decision behind it. ARSA’s explainer on face detection, recognition and verification sets out which of those tasks a check-in needs.
| Item | Published Value |
|---|---|
| Image endpoint latency | Hundreds of milliseconds |
| Active liveness latency | Several seconds, scaling with video length |
| Client timeout on video endpoints | 30 seconds minimum |
| Memory | 8 GB |
| Network for inference | None required |
| Developer licence | $1,490 per year, one machine, up to 1,000 identities |
| Node-locked licence | $5,900 per year, one machine, unlimited internal enrolments |
| Multi-node licence | $14,900 per year, high availability supported |
| Air-gapped or classified | Quoted after a site survey |
Two published figures matter for supervision. ARSA’s benchmark page gives the server model a 94.41% true accept rate at a false match rate of 1 in 10,000 on VGGFace2. At the same false match rate the edge model accepts 85.40%, which means roughly 1 in 7 legitimate attempts is rejected, against 1 in 18 for the server model. An agency that treats a rejection as a possible breach should pick the server model and plan a fallback route for people the system rejects.
The SDK leaves several jobs to the buyer. TLS ends at the buyer’s reverse proxy, CORS is open by default and must be restricted, and rate limiting is not built in. A remote check-in service that faces the internet needs all three before it goes live.
ARSA delivers outside Indonesia. List prices are in USD, invoicing is available in USD or EUR, and where visa conditions prevent ARSA engineers travelling, a qualified local partner installs under ARSA supervision. A feasibility assessment costs $4,500, and the fee is deducted from the project fee if the agency contracts within 90 days (ARSA services). Questions to bring to ARSA: how your check-in app should call active liveness, what video your phones will send, and which licence fits your caseload.
FAQ
Frequently Asked Questions
Do Community Corrections Agencies Use Liveness Detection For Remote Check-Ins?
Some agencies use face verification for remote check-ins, but the public sources reviewed here do not describe their liveness methods. England and Wales, Massachusetts and Queensland all match faces to confirm identity. None of the five sources explains how the system rejects a photo or a replayed video, so buyers should ask any vendor directly.
What Is The Difference Between Active And Passive Liveness Detection?
Passive liveness checks a single image for signs of a spoof, and active liveness asks the person to perform a randomised sequence of movements on video. Passive is invisible to the user and takes a few hundred milliseconds. Active takes several seconds and is the stronger check. ARSA’s SDK includes both in every licence.
How Does Passive Liveness Detection Work?
Passive liveness detection analyses one image for the signals that separate a live face from a printed photo, a screen replay or a mask. The person does nothing extra. That makes it a good fit for a fast check, with active liveness kept for events where proof matters more than speed.
Why Does On-Premise Liveness Detection Matter For Justice Agencies?
On-premise processing keeps biometric data under the agency’s control and adds no vendor to its data protection obligations. Under UK GDPR, Queensland’s Information Privacy Act and the CJIS Security Policy in the US, every extra party that holds the data is another one to audit. The licensed ARSA SDK runs with no outbound connection at inference time.
What Security And Privacy Standards Should A Face Recognition API Meet For This Use?
At minimum it should encrypt traffic, keep biometric payloads out of its logs, keep templates off the API, and separate each application’s data. ARSA’s SDK never persists request bodies, never exposes templates, and gives each API key its own isolated face database. The buyer adds TLS, restricts CORS and adds rate limiting at a proxy.
CTA
Build Your Check-In Service On Infrastructure You Control
Run face verification and active liveness for remote check-ins inside your own network, with no outbound connection at inference time from the licensed product. Read the specifications, the pricing and the free demo terms on the Face Recognition and Liveness SDK page, or talk to an engineer about your caseload.