Western Australia Police end five-month live facial recognition trial: 1.38 million faces scanned, 122 arrests, nine false alerts
Edited by Elliot Marsh, Editor · written with AI from the linked sources (how)
WA Police concluded a five-month trial of real-time facial recognition cameras on marked vans, with 100 deployments at 43 locations (including the Perth Royal Show). Police report 1,380,882 faces scanned, 122 arrests, nine incorrect alerts (stated error rate 0.00065%), engagement with 146 registered sex offenders and 10 welfare checks. The vans have been withdrawn from public deployment while police run an internal evaluation of performance, safeguards and community impact. No decision on permanent rollout is reported.
Entities: Western Australia Police, Col Blanch, Civil Liberties Australia, Bill Rowlings, Office of the Information Commissioner WA, The Guardian
0 primary
What happened
Western Australia Police ended a five-month trial of live facial recognition cameras mounted on marked vans. There were 100 deployments at 43 locations, including the Perth Royal Show. Police say 1,380,882 faces were scanned, leading to 122 arrests, nine incorrect alerts, engagement with 146 registered sex offenders and 10 welfare checks. The vans are now off public deployment while police run an internal review of performance, safeguards and community impact, and no decision on permanent rollout has been reported.
Why it matters
This is a rare set of operational numbers from a real public deployment of live facial recognition in Australia, so it will likely shape the argument over whether WA keeps the technology. People in public places such as shows and busy streets were scanned without suspicion, which puts privacy, bias and oversight questions in front of regulators and the state government. The figures are police-reported and the review is internal, so they do not settle whether the system is accurate or proportionate. Nothing changes in practice until a rollout decision is made.
What is noise
The 0.00065% error rate is nine false alerts divided by all faces scanned, which flatters the system. A fairer test is how many of the alerts were wrong, and the coverage gives no total alert count for that. The headline "one million faces" also mixes scans with people, because the same person may be scanned more than once. Without independent data, claims that the trial "worked" or that most people were "unaffected" rest on police framing, and the 122 arrests are not broken down by offence or by whether they led to charges.
Watch next
- 01Whether WA Police publish the internal evaluation, and whether it includes total alerts, the share that were correct, charges laid, and any breakdown by demographic group.
- 02Any decision on permanent or expanded deployment, and whether it comes with new legislation, a code of practice or an independent oversight body.
- 03Statements or reviews from the WA Information Commissioner and Civil Liberties Australia, and whether other Australian police forces cite these results to start their own trials.
Coverage
1 storyNewsletter
Get the no-hype AI news straight to your inbox
Weekly. No spam, unsubscribe any time. Privacy
More adoption signals
Full feed →- Meta releases Muse Spark Open, an open-weight ~816B-parameter multimodal mixture-of-experts model, with Hugging Face support15 Oct 202680
- Google DeepMind launches AlphaGenome Atlas, a free public database of predicted effects for 9 billion possible human genome variants8 Sept 202679
- Apple ships Gemini-powered Siri beta with iOS 27, excluding EU and China at launch15 Sept 202678
- Anthropic releases Claude Haiku 5.5 at sharply lower prices, halves Sonnet 5.5 cache read cost, and adds monthly API credits for subscribers7 Oct 202677