This Is Surveillance

Most organisations deploying video analytics already have cameras. The assumption is that adding analysis is a technical upgrade. It is not.

What changes

A recording camera is passive. Someone reviews footage after an event. Most footage is never watched.

An analysing camera observes everyone continuously and produces structured records about their behaviour.

Which means the practical reality changes even if the hardware does not: what was theoretically reviewable becomes actually analysed, at scale, all the time.

The existing notice does not cover it. Signage saying the area is under video surveillance does not describe automated analysis, and where the analysis produces personal data the notice obligation is fresh.

What to settle before deployment

What is analysed, specifically: which detections, which zones, which hours.

What is not. The negative list matters more than the positive one for anyone reading it.

Whether anyone is identified. This is the question everyone asks and the one most notices avoid.

What is retained: frames, derived detections, or aggregates only, and for how long.

Who can see it, under what trigger.

Exclusions: welfare facilities, changing areas, medical rooms, break areas — enforced in the camera configuration rather than in the reviewing software.

What it will not be used for, stated and enforced.

The workplace case specifically

Where the power imbalance makes this hardest.

Consent from an employee is generally not a sound legal basis in jurisdictions where consent must be freely given, because it cannot be freely refused.

Which means the justification must rest on something else, and that something else has to be documented rather than assumed.

Consultation is a legal requirement in several jurisdictions and improves the design everywhere: the people doing the work know which detections will be wrong and which workarounds exist.

Purpose limitation is the boundary most often crossed. Safety detection repurposed for productivity assessment produces exactly what it does everywhere else — people avoid the camera, and the safety data degrades.

The public space case

Notice must be visible before someone enters the analysed area, not at the till.

Vantage over adjacent property — a neighbour's garden, a public pavement — is a separate issue and frequently the one that generates a complaint.

Mask what you do not need to see, in the camera configuration, and record that you did.

The question to answer honestly

"Are you watching me?"

The truthful answer for a well-designed counting system is: the system counts people and keeps no record of individuals.

For a system doing identification it is yes, and saying anything else is the kind of thing that surfaces later at maximum cost.

Design so that the honest answer is the reassuring one, which is almost always achievable and is the whole argument for aggregating at source.

The question people actually ask

Design so the honest answer is the reassuring one.

"Are you watching me?"

For a well-designed counting system: the system counts people and keeps no record of individuals. True, checkable, and it ends most objections.

For an identification system: yes, and any other answer surfaces later at maximum cost.

If the honest answer is uncomfortable, the design is the problem rather than the wording, which is the whole argument for aggregating at source.

Test the notice against this question before publishing it.

Adding analysis to an existing estate

The most common route into this field and the one most often treated as a technical upgrade.

The cameras were installed and notified for recording.

Adding analysis changes what is happening: everyone in view is now observed continuously and structured records are produced.

Which requires a fresh assessment, a fresh notice and, for employees, fresh consultation.

"The cameras were already there" is not an answer to any of the three.

Treat it as a new deployment in the register, with its own entry, because that is what it is.