Ethics of Surveillance: Facial Recognition, Proportionality and the Chilling Effect
IssueAugust 2026
PaperGS Paper 4
SubjectEthics and Ethical Crisis
Read time5 min
Context
The disclosure, in a Delhi Police affidavit before the Supreme Court in August 2026, that facial recognition technology and AI-enabled surveillance were deployed during protests raises a question that no amount of technical safeguarding resolves: whether identifying every participant in a lawful public assembly is a legitimate exercise of state power at all.
The Core Ethical Tension
Legality is not Sufficiency: a police force may lawfully photograph a public gathering, and individuals in public spaces have historically had limited expectation of anonymity. What changes with automated facial recognition is scale and permanence — the capacity to convert a crowd into a searchable list, retained indefinitely, and matched against every future appearance. An act that is trivial when performed once by a human observer becomes qualitatively different when performed continuously by a machine on everyone.
Security against Liberty is the Wrong Frame: posing this as a trade-off assumes both sides are commensurable and that more surveillance reliably produces more security. Where surveillance is directed at lawful protest rather than at crime, no security gain is even claimed — what is being protected is the state's informational advantage over its critics.
The Chilling Effect: the deepest harm is not to those identified but to those who, knowing they may be, choose not to attend. That harm is invisible, unmeasurable and borne by the political system as a whole — which is precisely why it is systematically underweighted in official justification.
Key Ethical Concepts and Applications
Proportionality (Constitutional and Moral): the doctrine, articulated in Indian constitutional jurisprudence in K.S. Puttaswamy (2017), requires that a measure restricting a right pursue a legitimate aim, be rationally connected to it, be the least restrictive means available, and be balanced in its effects. Mass facial recognition at a protest fails the third limb decisively — targeted identification of individuals suspected of specific offences achieves any legitimate policing aim without the blanket alternative.
Privacy as a Precondition, not a Preference: Puttaswamy grounded privacy in dignity and autonomy under Article 21, not in secrecy. On this account privacy is not something one gives up by entering public space; it is the condition under which free choice about association and expression is possible at all. Surveillance of assembly therefore burdens Article 19(1)(a) and 19(1)(b) as much as Article 21.
Foucault's Panopticon: Michel Foucault's analysis of Bentham's panopticon showed that the disciplinary power of surveillance lies not in constant observation but in the possibility of it — the observed internalise the watching and regulate themselves. Applied here, the state need not actually identify anyone for the effect to operate; the announcement that it can is sufficient.
Function Creep and the Ethics of Capability: systems built for one purpose migrate to others, because the marginal cost of an additional use is near zero once the capability exists. The ethical assessment therefore cannot be limited to the stated purpose; it must extend to every use the capability makes possible under a future administration with different intentions.
Algorithmic Bias and Distributive Harm: facial recognition systems have documented differential error rates across skin tone, gender and age. In a policing context a false match is not a statistical inconvenience but a wrongful detention, and the burden of that error falls disproportionately on populations already over-policed — converting a technical limitation into a distributive injustice.
Accountability and the Diffusion of Responsibility: when an algorithm produces the match, the officer acts on the output, and the vendor supplied the model, responsibility can be disclaimed at every step. Ethical deployment requires that a named human authority own each decision, and that the reasoning be reconstructible after the fact.
Institutional Challenges
India's Digital Personal Data Protection Act, 2023 provides for broad exemptions in favour of state instrumentalities, so the strongest constraints on state surveillance are constitutional rather than statutory — which means they are litigated after the fact rather than enforced in advance.
No dedicated statutory framework governs the police use of facial recognition: procurement, retention periods, accuracy thresholds, audit and deletion are largely matters of internal executive decision.
Oversight bodies lack the technical capacity to audit systems they are meant to supervise, so scrutiny defaults to accepting the deploying agency's own account of accuracy.
Procurement from private vendors introduces commercial confidentiality claims that obstruct independent evaluation of the very systems used in public enforcement.
Way Forward
Require prior statutory authorisation for any biometric surveillance of public assembly, with judicial or independent authorisation for specific deployments rather than standing permission.
Mandate published accuracy audits disaggregated by demographic group, and a defined maximum retention period with verifiable deletion.
Prohibit the use of facial recognition for the identification of participants in lawful assembly as a category, distinct from its use in investigating specific offences.
Establish an independent technical oversight body with the capacity and legal access to audit deployed systems, and place the burden of demonstrating necessity on the deploying agency.
Conclusion
The question raised by facial recognition at protests is not whether the technology works but whether a democratic state should possess, over its own citizens exercising their most basic political right, a capability whose principal effect is to make them think twice. Proportionality answers that question: where a less restrictive means achieves the legitimate aim, the more restrictive one is not merely excessive but unjustified — and a capability that chills lawful dissent is not made acceptable by the assurance that it will be used well.