Consent and surveillance — the line · AI, Ethics, and Society

When you tapped Agree to install Khalti, you consented to forty-two pages of terms drafted by a lawyer you will never meet, in a register of English you would not use in conversation, governing data flows you cannot inspect. You scrolled because the alternative was not being able to send money to your mother. The economists call this a contract. The lawyers call it a contract. Most Nepalis, if pressed honestly, would call it what it is — a fee for entry, paid in agreement to whatever the document happened to say. The question this section opens is whether the ceremony of clicking is the same thing as the substance of consenting, and what that distinction implies for the surveillance systems Nepali municipalities are right now signing contracts to deploy.

The legal idea of consent rests on four conditions, each of which has to be present for the agreement to mean anything. The person must be informed — they must actually know what they are agreeing to, in language they understand. The agreement must be specific — to a defined purpose, not to “any future use we think of.” It must be revocable — the person must be able to walk it back without losing what they already received. And it must be uncoerced — the alternative to consenting cannot be exclusion from things they need.

By those four tests, almost no Nepali digital consent is meaningful. The terms are not in plain Nepali. They cover not the present use but every possible future use. Revocation, if it exists at all, requires emailing an address that does not reply. And the alternative to agreeing is being shut out of the wallet, the bank app, the school portal, the ride app — the things that have become functionally compulsory in urban life. What we have, instead of consent, is ceremonial consent: a ritual whose function is to transfer legal liability from the company to the user, not to communicate anything to the user about what is happening to them.

This matters because the doctrine of consent is doing most of the heavy lifting in Nepali data law. The state’s argument for why a wallet, a hospital app, or a school platform can collect what it collects is, in almost every case, “the user agreed.” If that agreement is ceremonial rather than meaningful, then the doctrine is hollow — and the protection it appears to offer is a fiction.

The surveillance slope

Surveillance technology does not arrive announcing itself as surveillance. It arrives as something else — usually safety. The first Kathmandu Metropolitan City CCTV cameras went up around Ratna Park and Sundhara in the mid-2010s with the stated purpose of deterring pickpockets and helping the traffic police. Few people objected, because few people thought there was anything wrong with a camera on a pole.

The cameras were then upgraded to higher-resolution sensors. The video was then centralised in a control room. The control room then acquired a vendor pitch for face-recognition software — initially framed as a way to find missing children, then as a way to identify wanted persons, then as a way to count crowd density at protests, then as a way to “understand traffic flow patterns” using anonymised tracks that, with enough cameras, are not really anonymous at all. At each step, the stated purpose was reasonable. At each step, the previous step’s infrastructure made the next step almost free. By the time anyone asks whether the metropolitan city should have a live, persistent, face-indexed record of every adult who walks through Ratna Park, the cameras are already there, the budget is already spent, and the question is moot.

This is the surveillance slope, and it is not unique to Nepal. What is unique to Nepal is that there is no national framework — no equivalent of a data-protection authority, no equivalent of judicial oversight of mass surveillance, no equivalent of a public-interest litigation history strong enough to act as a brake — to slow the slope down. Each municipality is making the decision on its own, on the strength of a vendor pitch and a budget line.

Four words that hold the line

There is a usable line between legitimate safety use of surveillance technology and overreach, and it can be stated in four words. Each is a constraint. Together, they are the difference between a tool and a regime.

Proportionality. The intrusion must be the least intrusive available means of achieving the stated purpose. If the purpose is “reduce theft at Ratna Park,” more lighting and a beat officer may achieve more reduction with less data than a face-recognition system. If a less intrusive option works, the more intrusive one is, by definition, disproportionate.

Purpose-binding. Data collected for purpose X cannot be used, without fresh authority, for purpose Y. Footage collected to catch pickpockets cannot, six months later, be re-purposed to identify participants at a peaceful rally. The data does not become available for any purpose just because the camera already exists.

Sunset clauses. The authority to collect — and the retention of what has been collected — must expire on a defined date, and renewal must require a fresh public justification. A surveillance programme without a sunset clause is a permanent regime. With one, it is a time-bound experiment that must prove its value to continue.

Independent oversight. Some body other than the agency doing the collecting must have the standing, the access, and the actual power to say no — to demand audits, to investigate complaints, to publish findings, to order the deletion of unlawfully retained data. Without an outside referee, every internal rule is a suggestion.

A worked example — the metro CCTV proposal

Imagine, concretely, that Kathmandu Metropolitan City puts out a tender in 2026 for a city-wide CCTV network with integrated face-recognition, pitched on the basis of public safety. Apply the four-word test.

Proportionality asks: has the city demonstrated that less intrusive interventions — better lighting, more foot patrols, targeted enforcement at known hotspots — have been tried and found insufficient? If not, the proposal fails at the first step. Purpose-binding asks: is the face-recognition use restricted, in writing, to a defined set of purposes (missing-person searches, identification of suspects in active investigations), with criminal penalties for off-purpose use? If the use is “general public safety,” the purpose is unbounded and the binding is fictional. Sunset asks: does the authority lapse after, say, three years, requiring the metropolitan council to publicly review the programme’s actual effects before renewing? Or is it perpetual? Oversight asks: who, outside the metropolitan office and the vendor, can audit the system, investigate complaints, and order deletions? If the answer is “no one,” the system is, in operational terms, accountable only to itself.

A proposal that satisfies all four is at least defensible. A proposal that satisfies none — and most current vendor proposals satisfy none — is a blank cheque written in citizens’ faces, signed before anyone has read it.

Check your understanding

Quick check

—

Kathmandu Metropolitan City is evaluating a city-wide CCTV-with-face-recognition proposal pitched as a public-safety measure. Which framing best separates a legitimate, defensible deployment from an overreach?

If the technology is modern and the vendor is reputable, the deployment is defensible.
If the stated purpose is public safety, the deployment is defensible regardless of scope or oversight.
The deployment is defensible only if it satisfies proportionality (least-intrusive means), purpose-binding (defined uses, no drift), sunset (time-bound authority requiring renewal), and independent oversight (an outside body that can audit and order deletions).
Any CCTV system that includes face-recognition is automatically illegitimate and must be banned.