Auror says its software is a tool to keep frontline workers and communities safer. File photo. Photo: 123RF
Police say they are checking how they use a system that identifies number plates and combats retail crime.
Officers accessed the system run by Auckland company Auror over 200,000 times last year.
The footage from carparks and inside stores is regularly used in shoplifting and other prosecutions, but has been subject to several failed challenges in court.
"Police regularly reviews its use of technologies and capabilities. For example, an unrelated Auror assurance review has been underway since mid-2024," police said in a statement.
They also completed an audit of their use of private automated number plate recognition (ANPR) platforms at the end of last year and are considering its results. Auror runs the largest ANPR platform.
Police recently [https://www.rnz.co.nz/news/national/539759/no-charges-after-golriz-ghahraman-pak-n-save-shoplifting-allegation
decided not to file charges] after a shoplifting allegation was made against former Green MP Golriz Ghahraman, which arose at a supermarket using the Auror platform.
Auror last week said it was "disheartening" that it was being mischaracterised as a surveillance system.
"Newsflash: We aren't a camera company and we're not a surveillance network," chief executive Phil Thomson said online.
The software was a tool to keep frontline workers, customers and communities safer, he said.
The cameras linked to the software are privately owned by shops, councils and the likes.
The Police Association said new technology like Auror's should not be feared, as it helped boost strained policing resources - and should be extended to greater use of facial recognition.
The union newsletter at the same time noted Auror's new partnership with giant US taser-and-bodycam maker Axon.
"NZ Police and the government need to be open to doing business with such partnerships to leverage the best technology to enhance the safety of Kiwis," association president Chris Cahill said in the newsletter.
Police have had contracts with Auror and Axon for years already.
They use other private tech firms for what appear to be fairly limited facial recognition operations - internal figures show police used the tech about 90 times within one of their main biometric systems from early 2022 to mid-late 2024, and only ever on still images, never on live video feeds; that remains banned under the police's first policy on facial recognition released last year.
But Cahill wanted more than this.
"Facial recognition technology (FRT) is an area where Aotearoa needs to break the shackles of yesterday's thinking or false understanding of the capacity of technology," he wrote.
"It is ridiculous that supermarkets and other retailers are leading the charge with FRT while police sits back and worries about outdated views on what, in the era of saturation social media and CCTV, constitutes privacy. Police needs to focus on the safety benefits FRT can bring to communities.
"It can be used ethically, and we need to dispel incorrect reporting on how well algorithms work. Our courts and public watchdogs also need to move with the times."
For years, a lot of attention has been paid globally to research in 2019 in the United States that showed racial bias in facial recognition algorithms that were more prone to misidentify brown and black people, and women, than white men.
However, leading systems have become much more accurate in recent years, according to benchmark US testing.
Japanese tech company NEC, which police have been using, said on its website it used advanced AI algorithms "to mitigate biases inherent in facial recognition" associated with factors such as age, gender, and ethnicity.
The first tests done on New Zealanders' faces of FRT built into a new Department of Internal Affairs online recognition system, last year showed no appreciable bias, though the sample size was small at less than 200 faces.
Sign up for Ngā Pitopito Kōrero, a daily newsletter curated by our editors and delivered straight to your inbox every weekday.
