How accurate is WA police facial recognition after 1.38 million scans?
A five-month trial in Western Australia yielded 122 arrests and nine false alerts, prompting debate over how error rates should be calculated for public safety AI.
Key points
- Western Australia police scanned 1,380,882 faces during a five-month trial of real-time facial recognition technology.
- The deployments resulted in 122 arrests, nine incorrect alerts and engagements with 146 registered sex offenders.
- Civil liberties advocates argue the official error rate ignores the legal outcome of the arrests made.
The Guardian reported that a five-month trial of real-time facial recognition in Western Australia concluded with 122 arrests and nine false alerts.reported Police deployed surveillance vans across 43 regional and metropolitan locations, scanning more than 1.38 million faces.reportedA recent trial in Western Australia used facial recognition on over a million people, leading to 122 arrests and nine mistakes.The WA police trial processed 1.38 million face scans with a reported 0.00065% false positive rate, though critics question the metric's validity. The trial ended on a Saturday, providing a dataset for evaluating the technology's operational performance in public spaces.
Evidence: reported
Police use facial recognition to scan one million faces in WA trial leading to 122 arrestsaifleet-guardian
Evidence: reported
Police use facial recognition to scan one million faces in WA trial leading to 122 arrestsaifleet-guardian
The announcement
WA Police stated the system generated nine incorrect alerts against the total volume of scans. Commissioner Col Blanch addressed the false matches at a media conference, confirming officers engaged with each individual flagged incorrectly. He emphasized that no complaints were filed and that the error count remained very low, though not zero.
The context
The deployments included three operations at the Perth Royal Show, where 223,749 faces were scanned.reported These specific events resulted in 15 alerts, three arrests and seven interactions with registered sex offenders.reported Overall, the technology led officers to engage with 146 registered sex offenders and locate 10 people for welfare checks during the trial period.reported
Evidence: reported
Police use facial recognition to scan one million faces in WA trial leading to 122 arrestsaifleet-guardian
Evidence: reported
Police use facial recognition to scan one million faces in WA trial leading to 122 arrestsaifleet-guardian
Evidence: reported
Police use facial recognition to scan one million faces in WA trial leading to 122 arrestsaifleet-guardian
Reading between the lines
Commissioner Blanch acknowledged the technology was not perfect but maintained that officers followed up on alerts similarly to how they would handle public suspicions. He stated the goal was community engagement rather than immediate arrest for every alert. This approach frames the false positives as manageable interactions rather than systemic failures in the identification process.
The business case
The low reported error rate of 0.00065% suggests high precision in matching, which is critical for vendor selection in surveillance markets.reported However, the debate over calculating errors based on convictions rather than alerts highlights the risk of reputational damage if legal outcomes do not support the initial matches.
Evidence: reported
Police use facial recognition to scan one million faces in WA trial leading to 122 arrestsaifleet-guardian
Why it matters
- Platform architects must consider how error rates are defined, as operational metrics may differ from legal accuracy standards.
- The volume of scans indicates the infrastructure requirements for processing real-time video feeds in high-density public areas.
- Vendor claims of low error rates may face scrutiny from civil liberties groups who demand conviction-based accuracy metrics.
- The trial shows that even low error rates result in hundreds of engagements, impacting resource allocation for follow-up investigations.
Topics: facial recognition, surveillance, privacy, law enforcement, AI ethics
Updated
Then / now / next
Facial recognition trials were smaller in scale and less documented in public reports.
A large-scale trial has provided concrete data on scans, arrests and false alerts.
Civil liberties advocates will monitor court outcomes to determine the true accuracy rate.
Story timeline
- Then Facial recognition trials were smaller in scale and less documented in public reports.
- 02 Nov 2026 Police use facial recognition to scan one million faces in WA trial leading to 122 arrests
- Now A large-scale trial has provided concrete data on scans, arrests and false alerts.
- Next Civil liberties advocates will monitor court outcomes to determine the true accuracy rate.
The strongest doubt
Counterpoint from the sources
Civil Liberties Australia chief Bill Rowlings argued the police were exaggerating the success by using a flawed error metric that ignores whether arrests led to convictions.
For readers
Policy
- The definition of error rates is contested between police and civil liberties advocates.
Researchers
- The trial provides a large dataset of 1.38 million scans for analysis.
Business
- Vendors may face pressure to report accuracy based on legal outcomes rather than system matches.
Ask the article
Local keyword search of this article; no generated answers.Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Questions readers ask
What was the error rate of the facial recognition trial?
WA Police reported an error rate of 0.00065%, based on nine incorrect alerts out of 1.38 million scans.
How many arrests resulted from the trial?
The trial resulted in 122 arrests, alongside engagements with 146 registered sex offenders.
Did anyone complain about the false alerts?
No complaints were filed regarding the nine incorrect alerts during the trial period.
Connected stories
Reading receipt
- Western Australia police scanned 1,380,882 faces during a five-month trial of real-time facial recognition technology.
- The deployments resulted in 122 arrests, nine incorrect alerts and engagements with 146 registered sex offenders.
- Civil liberties advocates argue the official error rate ignores the legal outcome of the arrests made.
Still open: How many of the 122 arrests resulted in convictions?
Still open: Will the error rate definition change for future deployments?