Western Australia police reported nine incorrect alerts after scanning 1.38 million faces.
Why it matters: The figures will inform debates over police use of biometric surveillance, privacy safeguards and accountability for false matches. But the scan-based calculation does not show the system's alert-level accuracy or whether reported arrests and other contacts resulted from facial-recognition alerts.
- The trial ran from June 22 through October 31, 2026, across 100 deployments at 43 locations, according to <a href="https://www.theguardian.com/australia-news/2026/nov/02/wa-police-facial-recognition-technology-trial">The Guardian's report</a>.
- Police reported 1,380,882 face scans, 122 arrests, 146 engagements with registered sex offenders and 10 welfare checks; available reporting does not establish that those events were caused by facial-recognition alerts.
- Police reported nine incorrect alerts. That figure equals 0.00065% of all face scans, not the system's error rate among alerts; no overall alert total is reported in the available material.
- WA Police says every alert requires officer review and is not, by itself, grounds for arrest or other police powers, according to its <a href="https://www.wa.gov.au/organisation/western-australia-police-force/live-facial-recognition-technology">official guidance</a>.
Western Australia Police used marked surveillance vans and cameras in what ABC News described as Australia's first overt live facial-recognition trial. The deployments compared live camera feeds with a police watchlist - a list of people whose images police wanted the system to check - using NEC's NeoFace M40.
The trial ran from June 22 through October 31, 2026, or 132 calendar days inclusive, across 100 deployments at 43 locations. It scanned 1,380,882 faces. The reports also list 122 arrests, 146 engagements with people registered as sex offenders and 10 welfare checks. In plain terms, the latter figures refer to police contacts with people on the sex-offender register and contacts intended to check someone's safety. The available reports do not say whether these events were triggered by facial-recognition alerts or simply occurred during deployments.
At the Perth Royal Show, three deployments scanned 223,749 faces, produced 15 alerts and led to three arrests, according to The Guardian. That is a site-specific figure, not a reported total for the entire trial.
Police reported nine incorrect alerts. Commissioner Col Blanch called that “a very, very, very low number,” while acknowledging, “It's not zero. We'd love it to be zero.” Police said no one was arrested as a result of those alerts and that officers interacted with each person involved.
Dividing nine incorrect alerts by all 1,380,882 face scans produces 0.00065%. That is a scan-based ratio, not an alert-level error rate. The available reports do not provide the total number of alerts, demographic error rates or a reliable link between alerts and later arrests, charges or other enforcement outcomes. They also do not report a final evaluation.
WA Police says non-alerted faces are pixelated, biometric data is automatically deleted and information is not shared with third parties. Its guidance says every alert must be reviewed by an officer. The Office of the Information Commissioner said it was not involved in implementing the trial and warned of potential privacy harms in its statement.
By the numbers
- 1,380,882 - faces scanned during the trial
- 9 - incorrect alerts reported by police, equal to 0.00065% of all scans
- 223,749 - faces scanned at three Perth Royal Show deployments, which produced 15 alerts and three arrests
Yes, but: The 0.00065% figure uses all face scans as its denominator. Without the total number of alerts, demographic breakdowns and verified links between alerts and enforcement outcomes, it cannot establish the system's alert-level accuracy or broader effects.