The cameras went live at Victoria. Nobody had to be watching for it to work.

The story

Transport for London is trialling facial recognition technology at stations in a bid to identify sex offenders, robbers and knife users. Cameras went live at Victoria this week, following a six-month trial at other stations. In that trial, 530,000 scans were carried out and produced no matches. TfL has not said how many further stations will be added, or what happens to the biometric data of the hundreds of thousands of passengers scanned who were never suspected of anything.

The reframe, with the manuscript

Half a million scans and no matches sounds, on the surface, like a system that has not yet found anything to justify itself. Read the other way, it is a system doing exactly what it was built to do. Facial recognition at a train station does not exist to catch the person it is looking for. It exists to look at everyone, continuously, in case that person eventually walks through. The absence of a match after 530,000 attempts is not a failure rate. It is the ordinary output of blanket scanning, and it will be reported that way for as long as it takes for the first match to arrive.

What makes this different from a police officer standing at the barrier is precisely the thing that makes it easy to wave through. Nobody at Victoria this week was watched by a person. Nobody was stopped, questioned, or even aware, in most cases, that their face had been checked against a database. The manuscript's account of how modern institutions have moved past the need for a watching human describes this shift directly:

Modern institutions increasingly do not need to observe individuals directly in order to understand them. Participation creates visibility automatically. The system does not need to ask. The data already exists.

Riding the Tube is about as involuntary a piece of daily participation as exists in London. There is no meaningful way to opt out of walking past a camera at Victoria and still get to work. That is the condition the manuscript is describing when it distinguishes this kind of visibility from surveillance in the older, more familiar sense, someone watching, someone following, someone deciding to look:

The issue is not surveillance in its traditional sense. It is legibility.

Legibility does not require suspicion. It requires only that a face cross the frame. TfL's own six-month figure makes the scale of that plain: 530,000 ordinary journeys, run through matching software, on the strength of nobody in particular. The trial's stated purpose, catching sex offenders, robbers and knife users, is a real and defensible one. It is also the purpose that makes the scanning of everyone else easiest to justify, because the people who object to being scanned on their commute are rarely the people the system says it is looking for, and objecting looks, to anyone watching the exchange, like exactly the kind of behaviour worth a second look.

None of the half a million people scanned at those earlier stations were told, station by station, that this was happening to them specifically. They were told, in the way TfL announcements are typically told, that the technology existed, that a trial was under way, and that it was aimed at serious offenders. The manuscript's broader argument in this chapter is that this is now the normal shape of consent: blanket, general, agreed to in principle by a public that is rarely shown the specific instance in which it applies to them. A camera at Victoria does not ask permission from the commuter walking past it any more than a bank statement asks permission before recording a 3am transaction. Both simply generate the record. What differs is only whether anyone later has reason to look at it.