It began as an utterly ordinary errand. Matt Arnold, a 46-year-old comedy promoter, popped into his local Sainsbury’s in East Dulwich, London, to pick up some shopping and a bottle of alcohol. He moved through the familiar rhythm of the supermarket: selecting items, heading to the self-service checkout, scanning his groceries, and tapping his Nectar loyalty card as millions of shoppers do every day. There was nothing furtive about his behaviour, no reason for anyone to look twice at him. But then two members of management approached. They told him he had to leave. The reason, they explained, was that the store’s live facial recognition technology had flagged him in connection with an incident “earlier in the week.” Arnold was confused, then shocked. He had not stolen anything, argued with anyone, or done anything remotely suspicious. Yet here he was, being treated like a shoplifter in front of other customers, all because an algorithm had decided his face matched someone on a watchlist. As he later put it, with a mixture of disbelief and frustration: “A shoplifter does not walk around with that much shopping, they don’t scan it through, they don’t put their Nectar card through.” The absurdity of the accusation was obvious to anyone using basic human judgment. But in that moment, the human judgment of the staff seemed to have been overridden by the authority of a machine.
Arnold described the experience as a “terrifying glimpse of the future,” and it is not hard to understand why. After being confronted, he was escorted out of the store, publicly and without any real chance to defend himself. As he left, he noticed an overhead CCTV screen displaying his face inside a red circle, a visual marker that made him look like a wanted criminal. It was a deeply dehumanising image: a person reduced to a target, framed by surveillance technology, with no context, no history, no benefit of the doubt. What upset him most, he said, was the thought that this is what the future could be, where people simply listen to what the machine tells them to do without thinking about the consequences. That fear is not abstract. It is already happening. The staff members who stopped him were not acting out of malice. They were likely following a protocol, trusting a system that had been presented to them as highly accurate and reliable. But in doing so, they abandoned the most basic part of their own judgement. They saw a man buying groceries, scanning his own items, using his loyalty card, and still treated him as a threat because a screen told them to. That is the real danger of facial recognition technology: not just that it makes mistakes, but that it makes people stop thinking for themselves.
In the days that followed, Sainsbury’s head office contacted Arnold to apologise. The supermarket said the incident was caused by human error, not by the facial recognition technology itself, and that it had paused the use of the system at the East Dulwich store while it investigated. The technology supplier, Facewatch, also defended its product, claiming a 99.98 percent accuracy rate and insisting that every match is reviewed by a trained manager. A spokesperson said that a correct alert had been sent to the retailer, but that it was subsequently mishandled in store. For Arnold, however, this explanation was not entirely reassuring. He pointed out that there are likely more people who have been caught out by this system, and that mistakes are inevitable when staff are not properly equipped to implement the AI. His suspicion is supported by another recent case: Warren Rajah, a Sainsbury’s customer in Elephant and Castle, was also wrongly challenged for shoplifting after a facial recognition alert. In both cases, the technology did not directly accuse anyone; it flagged a possible match, and human employees acted on that flag. But that distinction, between the algorithm’s error and the human’s error, matters little to the person who has been humiliated in public. The system is designed to identify people, and when it identifies the wrong person, the consequences fall on them.
This incident is part of a much larger pattern of concern about facial recognition technology in public spaces. Earlier this year, British Transport Police revealed that out of 330,000 commuters scanned with facial recognition cameras in London, only one person had been flagged as a criminal, and even that was a false alarm. That statistic is staggering: 330,000 innocent people were surveilled, their faces captured and compared against databases, all in the hope of finding someone who was not even there. Silkie Carlo, the director of the civil liberties group Big Brother Watch, has called on Sainsbury’s to scrap the technology altogether, arguing that it is treating customers like criminals. She warned that serious mistakes like the one involving Arnold are inevitable when a national retailer carries out hundreds of thousands of indiscriminate ID checks with such sinister surveillance technology. Her point is not just about privacy, though that is important enough. It is about the fundamental relationship between a shop and its customers. A supermarket should be a place of trust and routine, not a place where shoppers are scanned, analysed, and judged by machines before they even reach the checkout. When people go to buy their groceries, they should not have to worry about being accused of a crime they did not commit.
The human cost of this technology is often lost in discussions of accuracy rates and error margins. But consider what it actually feels like to be in Arnold’s position. You are going about your day, perhaps thinking about dinner plans or a work deadline, when suddenly you are confronted by authority figures who tell you that a machine has identified you as a criminal. You are not given a chance to explain, or at least not a meaningful one. You are told to leave, and as you walk out, you see your own face on a screen, circled in red, like a character in a police drama. The shame, the confusion, the helplessness, all of it is real, and it lingers long after the apology arrives. Even if the technology is 99.98 percent accurate, that still means two in every ten thousand scans are wrong. When a system is used on hundreds of thousands of shoppers, those tiny percentages translate into real people, real mistakes, and real damage. And because the technology is so new, and so opaque, most people do not even know they are being watched. They have no idea that their face is being scanned as they browse the aisles, no way to challenge the system, and no clear path to justice when it fails them. The burden is placed entirely on the customer, who is expected to prove their innocence after being accused by an invisible algorithm.
Arnold’s story should serve as a warning, not just to Sainsbury’s, but to every company and institution considering the use of facial recognition technology. The problem is not simply that machines make mistakes; it is that human beings are too willing to believe them. When a trained manager receives an alert from a system that has been marketed as nearly infallible, they are likely to assume it must be correct, even when all the evidence in front of them suggests otherwise. This is known as automation bias, and it is one of the most dangerous side effects of introducing AI into everyday decisions. The technology should be a tool that supports human judgement, not a substitute for it. But in practice, it often becomes the opposite: a silent authority that employees are afraid to question. Sainsbury’s has apologised and paused the technology at one store, but that is not enough. There needs to be a broader conversation about whether facial recognition should be used in retail at all, and if it is, what safeguards must be in place to protect innocent customers. People should not have to accept surveillance as the price of buying a sandwich. They should not have to worry that a machine might decide, without any basis, that they are a criminal. As Arnold said, this is a terrifying glimpse of the future. But it does not have to be the future we choose. We still have time to decide that human dignity matters more than algorithmic efficiency, and that no system should be allowed to treat ordinary people as suspects without their knowledge or consent.
