Sainsbury's pauses AI facial recognition after wrongly ejecting a shopper, again

[Cobra]

Sainsbury's has suspended live facial recognition technology at its East Dulwich store after Matt Arnold, 46, was wrongly flagged as a shoplifter and told to leave while waiting at a self service till for staff to approve an alcohol purchase on August 6. Arnold, a comedy promoter buying supplies for a nearby event, said he'd already scanned his shopping and Nectar card, and described the experience as a terrifying glimpse of the future, warning that staff can end up feeling pressured to follow an automated alert without applying their own judgment.

Sainsbury's apologised and said the mistake was caused by human error rather than a failure of the technology itself, which is supplied by Facewatch. The company said the system has a 99.98% accuracy rate and that every match gets reviewed by a trained manager before action is taken. Facewatch separately said a correct alert had been sent to the store but was then handled incorrectly by staff on the ground.

This isn't the first time this has happened at a Sainsbury's branch. In January, Warren Rajah was ordered out of a store in Elephant and Castle after being wrongly flagged by the same Facewatch system, with the security firm later confirming he wasn't even on its database at all, meaning staff had apparently targeted the wrong person following an unrelated alert. Rajah described the experience as Orwellian and said he was left wondering whether he now has to walk around fearful of being misidentified as a criminal.

Campaign group Big Brother Watch says this isn't an isolated pattern. With legal and policy officer Jasleen Chaggar saying the organisation regularly hears from members of the public left traumatised after being wrongly caught up in what she called privatised biometric surveillance, and arguing that government promises to regulate the technology amount to lip service unless the unchecked expansion by retailers actually gets reined in.

Sainsbury's says the wider rollout will continue regardless. Pointing to the technology's role in tackling theft and violence against retail staff, though Arnold has publicly questioned why it remains active at other branches given this is at least the second confirmed wrongful ejection tied to the exact same system at the exact same retailer

Sainsbury's branch pauses AI cameras use after shopper ejectedwww.bbc.co.uk

[Hitman04]

A stated 99.98% accuracy rate sounds reassuring until you remember that number still translates into real wrongful ejections at actual scale once you're running it across millions of transactions daily. Worth remembering

[RainyDayFund76]

The human error framing from Sainsbury's is doing a lot of convenient work here.

If staff keep making the same specific kind of mistake with the same specific system, at some point that stops being random human error and starts looking like a genuine design or training failure

[EdgeNode Joel]

Rajah not even being on the Facewatch database at all is the more alarming case of the two. That's not even a borderline match, that's staff acting on an alert meant for a completely different person entirely

[Calm Paige]

Big Brother Watch's point about unchecked expansion by retailers is worth taking seriously.

There doesn't seem to be much regulatory friction stopping any store from just rolling this out as long as they can point to a theft prevention justification

[Hyperdrive71]

Arnold's quote about staff following what the machine says without applying their own judgment is really the whole problem in one sentence. Automation bias is a real documented phenomenon and retail staff are not trained to second guess a system they're told to trust. That distinction matters a lot here

[Freya]

Continuing the wider rollout immediately after apologizing for the second wrongful ejection at the same company is a choice that's going to look bad regardless of the actual accuracy statistics being cited.

Worth remembering. That part surprised me

[Ridge47]

Wonder how many similar incidents go completely unreported because the person involved doesn't have the platform or inclination to actually go to the press about it.

These two cases becoming public might just be the visible tip of a much bigger pattern

[Saka31]

I think it will be less than 99

[Luca76]

Curious what the actual appeal or correction process looks like for someone wrongly flagged. Being confirmed innocent after the fact doesn't undo the public humiliation of being escorted out of a shop in front of other customers

[Anthony87]

Good roundup of a clearly important story. Facial recognition in everyday retail settings deserves way more public scrutiny than it currently gets given how normalized this rollout has become without much visible debate

[IronFist56]

The alcohol purchase detail is such a specific and almost absurd twist.

Being flagged as a shoplifter while literally standing there waiting for a staff member to approve a purchase is about as clear a false positive as it gets