Foodstuff’s FRT trial found compliant

The Office of the Privacy Commissioner (OPC) released in May the findings from its inquiry into Foodstuff North Island’s (FSNI) 2024 trial of facial recognition technology (FRT) as a deterrent to retail crime and violence towards staff and customers.
The overall finding from the inquiry was that the model deployed and progressively updated by FSNI during the FRT trial, complied with the Privacy Act. Key features of the compliant model included:
• Its clear and limited focus on identifying people who had committed serious harmful behaviour, including physical and verbal assault and higher value theft.
• Watchlists of people of interest were compiled by staff trained on the criteria for harmful behaviour. To be watchlisted, two staff members had to agree that the criteria had been met. Watchlist enrolment was set to expire after two years, and lists were not shared between stores.
• The elderly, those under 18 and people with known mental health conditions were not added to watchlists. Images not matching a store’s watchlist were immediately deleted, and matches not actioned were deleted by midnight the same day.
• Watchlist alerts needed to be generated by more than one camera, and checked by two trained staff members, before any intervention was made.
• Accuracy levels for watchlist matches were acceptable, with no apparent bias or discrimination in how watchlists were populated or in how decisions to intervene were made.
• Information gathered was not used for any other purpose. Access to the information was limited to authorised people only, with all access logged and regularly reviewed.
The report from the independent evaluator (Scarlatti) FSNI engaged, found evidence that FRT in trial stores had reduced serious harmful behaviour by an estimated 16%.

Intervention by supermarket staff approaching repeat offenders accounted for around half of this reduction, while the deterrent of FRT use stopping them returning, accounted for the remainder.
The OPC report also identified improvements that were needed, these included:
• Due to its invasive nature, FRT should be reserved for serious crime behaviour, such as targeted in the FNSI trial, and not for managing lower-level criminal behaviour, such as minor shoplifting or for people seen as difficult.
• For those under a trespass notice, these are issued for a variety of reasons and therefore need to be checked to ensure the criteria of harmful behaviour is met before they are added to a watchlist.
• Ongoing monitoring and review of FRT, to ensure watchlists do not contain out of date or biased information, and that data on misidentifications, including skin tone, is being captured to help understand and remedy these occurrences.
The OPC emphasised that their report “is not a green light for more general use of FRT”, but that it provides guidance for businesses considering FRT and what is required to employ FRT in a privacy protected way.