Foodstuffs facial recognition misidentifies Māori shopper at Rotorua New World

On 2024-04-02 a Māori woman shopping at New World Westend in Rotorua was approached by store staff and told she had been trespassed after a facial recognition alert from a Foodstuffs trial. The customer offered three forms of photo ID but was still asked to leave; Foodstuffs called it a genuine case of human error and said it reported the incident to the Office of the Privacy Commissioner. Experts and the Privacy Commissioner raised concerns about bias and accuracy in the trialled system, which was trained on international data and not specifically on New Zealand populations.

Foodstuffs (New Zealand) · Incident Apr 2, 2024 · Indexed Jun 10, 2026 · 3 sources

An inaccurate face match plus failed human verification caused staff to act on a false AI alert.
What
On 2024-04-02 a Māori woman shopping at New World Westend in Rotorua was approached by store staff and told she had been trespassed after a facial recognition alert from a Foodstuffs trial.
Incident date
Apr 2, 2024
Who
Foodstuffs (New Zealand)
Failure mode
Hallucination
AI surface
Computer Vision
Severity
Medium

What happened

A Māori customer (reported as Te Ani Solomon) shopping at New World Westend in Rotorua on 2024-04-02 was approached by two staff and told she had been trespassed after the store's facial recognition system flagged a match. She offered three forms of photo identification and told staff the image was not her, but staff proceeded to ask her to leave. Foodstuffs characterised the incident as a genuine case of human error, apologised and reported it to the Office of the Privacy Commissioner. The episode prompted public coverage and an inquiry by the Privacy Commissioner into the supermarket trial.

What broke inside the model

Failure path · mode profile · Hallucination
  1. 01 · TriggerA user asks for a fact, a citation, or a figure.
  2. 02 · Model stepThe model writes a fluent, confident answer.
  3. 03 · Control gapNothing ties the claim back to a real source.
  4. 04 · FailureA fabricated fact ships as if it were verified.
  5. 05 · ConsequenceThe false claim reaches a customer, a court, or the public.

Confidence holds, and even spikes, as the claim detaches from any source.

The facial recognition system produced a false positive by matching the customer's face to an image in the store's database of trespassed individuals; experts noted the system was trained on international datasets and may perform worse on Māori and Pasifika faces. Human verification procedures failed because staff relied on the AI match instead of their own judgement, so the automated alert triggered an incorrect human action. The combination of biased training data, false match risk for people of colour, and breakdown in human oversight caused the wrongful identification.

Public visibilityHigh
Regulatory exposureActive
Customer impactFew customers
Financial impactUnknown
Time to disclosureWeeks
  1. PressMāori woman mistaken as thief by supermarket AI not surprising, experts sayrnz.co.nz
  2. PressFacial recognition: Mum’s ‘horrible’ mistaken identity ordeal at New Worldteaonews.co.nz
  3. PrimaryInquiry into Foodstuffs North Island trial use of facial recognition (Office of the Privacy Commissioner)privacy.org.nz
Permalinkhttps://failureindex.ai/failures/foodstuffs-facial-recognition-misidentifies-ori-shopper
CitationAI Failure Index. "Foodstuffs facial recognition misidentifies Māori shopper at Rotorua New World" (FI-0490). Realm Labs. https://failureindex.ai/failures/foodstuffs-facial-recognition-misidentifies-ori-shopper (indexed Jun 10, 2026).
Share cardA branded image of this record for posts and slides.

Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0490. Full dataset at /data.

Note from Realm Labs, the Index steward

How Realm fits

Controls for this failure mode
  • Prism
  • OmniGuard
  • AI Detection & Response (AIDR)

This entry sits in the index's predictive wing: a system that scores, ranks, perceives, or steers rather than generates. Realm's runtime layer is built for the generative and agentic systems now moving into these same decision seats, where it watches a model's internal state and holds an unsupported claim or an unchecked action before it commits. The control gap on this record, an automated decision that reached people with no runtime check in front of it, is the same gap. The index keeps predictive failures on the record because the pattern carries straight into the systems shipping today.