The Person in Front

The people coming through the shop door behind me are not part of my shopping.

I haven’t selected them. I haven’t checked their references. I can just about account for the man in front. Me. I was hoping to make it to the milk without having to.

I mention this because one detail in a report about facial recognition has got underneath all the others. A woman was stopped after the system identified the people walking directly behind her.

Her face did not have to be mistaken for somebody else’s face. The trouble could travel the last few feet without that assistance.

Samuel Woodhams describes the incident in TechRadar’s October 29 investigation, which Joe Wilkins followed in Futurism on November 1. Woodhams attended a meeting in the House of Commons where shoppers described being wrongly accused of theft. One couple had gone to Home Bargains for vitamins. Another woman was accused at Budgens of stealing meat the previous week; she said she hadn’t visited for months. Some people had been stopped at the entrance, others at the tills. These were accounts of different failures, not a test establishing one defect in every camera.

The woman intercepted in front of the identified people interests me precisely because I cannot put her case into the comfortable version of this argument. Machine sees wrong face. Human believes machine. Improve machine, improve human, resume shopping.

We don’t know from the report whether the people behind her had been correctly matched. We don’t know what had put them in the system. But even if the identification had been perfect, somebody still had to connect the alert to the person being stopped.

I don’t have the footage or the employee’s explanation. The report doesn’t let me reconstruct that connection. It does make my confidence in arriving with the correct face look rather optimistic.

If somebody says I have stolen meat, I can say I haven’t. If he says I was here last week, I can try to establish where I was. These are miserable ways to spend an errand, but at least I have been given a proposition with my name on it.

What do I answer if the thing which brought him over wasn’t about me?

I could tell him my name correctly, show him my own face from several useful angles, and remain irrelevant to the event on his screen. Meanwhile I am the man talking too much near the door. Other customers could see that part without seeing the alert. I can do without an audience while I try to learn what I am denying.

Putting a person into this process is not, by itself, the missing safeguard. In the incident Woodhams reports, a person was already doing the stopping.

Facewatch, the technology provider, has an answer about responsibility. It told TechRadar that it takes responsibility for its technology and processes, while retailers are responsible for how their staff interact with customers. It said legal liability depends on the facts. It also said false alerts trigger a safeguarding process with the retailer to establish what happened, and challenged the impression that exceptional incidents represent everyday use across thousands of shops.

Those are material answers. A few accounts do not tell me how often this happens. A safeguarding process is not nothing. I would rather the two companies investigate a mistake than admire each other’s equipment over it.

But listen to what the customer has to locate between those answers. Was the wrong face matched? Was the right face matched to a bad record? Was somebody else selected to receive the accusation? The shopper doesn’t get to choose the failure. He may have to discover it before he can direct his objection to the people claiming responsibility for that part.

Facewatch and Sainsbury’s were invited to the Commons meeting and sent no representatives. Facewatch said it had concerns about the event’s balance, given Big Brother Watch’s campaigning, and had offered to answer questions afterwards. Sainsbury’s and Home Bargains didn’t answer TechRadar’s requests for comment; Booker Group, which owns the Budgens brand, declined to comment on record.

So the shoppers could describe what had happened to them. The room could not establish the path of every accusation. Being able to tell the story was not the same as being able to inspect it.

I haven’t inspected those systems either. I can’t clear a stranger’s record from here or tell the frightened shoppers that another branch will be better. I can’t even supply the sentence that would have got this woman past the intervention. Her own name? A request to look again? A finger pointed behind her? I don’t know what the employee had been shown.

It is an ugly limitation to discover while sitting safely outside the encounter. Inside it, I could be explaining myself with complete accuracy, growing louder, growing less convincing, trying to supply the missing fact from the only life I have information about.

I could lose the afternoon establishing that I was the wrong person to interrupt.

The missing information would still belong to the shop. So would the milk.

Original reporting: ‘Shaking from head to toe’ — How the facial recognition blame game leaves shoppers stranded (Samuel Woodhams, October 29, 2026).

Source: Innocent People Are Being Accused of Shoplifting by Faulty AI-Facial Recognition Systems at an Alarming Rate (Joe Wilkins, November 1, 2026).