For most of democratic history, the state could see a crowd but not remember it. A police officer watching a rally of 50,000 people could identify only a handful of faces and retain almost none of them by evening. This practical anonymity allowed citizens to participate in public life without fearing that their presence would become a permanent record.
Artificial intelligence has begun to dismantle that arrangement. During the recent youth-led demonstrations organised by the Cockroach Janta Party (CJP) at Delhi's Jantar Mantar, the Delhi Police deployed live facial recognition systems mounted on mobile command vehicles. The deployment has since become the subject of a Public Interest Litigation before the Delhi High Court, raising a significant constitutional question: Can the freedoms of assembly and dissent survive when public presence itself becomes data?
The controversy is not merely about cameras. Cameras have watched Indian crowds for decades. The fundamental transformation lies elsewhere—it is in the state's newfound ability to remember. A face captured today can potentially become an identity stored, matched and retrieved years later.
How does facial recognition technology work?
Facial recognition is often described as a smarter camera. That description is misleading. It is a biometric identification system that converts human features into a digital signature. When a camera captures a face, algorithms analyse facial landmarks—the distance between the eyes, the contour of the jaw, the depth of the nose bridge—and transform them into a numerical template known as a ‘faceprint’.
The system does not merely record. It interrogates. It continuously asks whether a face corresponds to an identity already present in a database.
Connected with police records and surveillance networks, facial recognition can scan thousands of individuals within minutes, without their knowledge or consent. Unlike fingerprints or iris scans, it requires neither cooperation nor physical contact. Presence in a public space becomes sufficient for identification.
How did the tech evolve?
Facial recognition's journey from research labs to policing has been gradual but inexorable. Early experiments in the 1960s involved manually plotting facial measurements. The US Department of Defense's FERET programme in the 1990s accelerated automated matching techniques, while security concerns after 9/11 pushed the technology into border control and law enforcement.
India's adoption followed a familiar trajectory of function creep—a process in which a technology introduced for one limited purpose gradually expands into much broader uses.
Delhi Police first acquired Automated Facial Recognition Software in 2018 following a Delhi High Court order aimed at tracing missing and trafficked children. Within a year, however, the same technology was reportedly being used for crowd surveillance during political gatherings at Ramlila Maidan. Between 2020 and 2022, facial recognition systems were deployed during anti-CAA protests and post-riot investigations. By 2026, systems such as Ikshana marked another shift—from analysing footage after an incident to real-time monitoring of crowds.
Digital rights groups argue that such expansion is neither accidental nor exceptional. Surveillance technologies often begin with socially unassailable objectives before moving into increasingly intrusive domains.
What surveillance technologies were deployed at the CJP protest?
At the CJP protest site, police deployed specialised mobile command and control vehicles equipped with:
360-degree high-definition cameras
Live facial recognition systems
Automated Number Plate Recognition (ANPR) technology
Reports also indicated the use of portable surveillance towers and wearable AI-enabled monitoring devices. Earlier RTI disclosures obtained by digital rights organisations revealed that Delhi Police treated matches above an 80% similarity threshold as positive identifications. However, technical experts caution that similarity does not equal certainty. Outdoor environments are inherently difficult for facial recognition. Changing light, movement, facial expressions, camera angles and crowd density all affect accuracy. An algorithmic match represents probability—not proof.
What does the Delhi Police say, and why are civil liberties groups concerned?
The Delhi Police argues that live surveillance is necessary to maintain public order, prevent violence and identify known offenders. Officials have maintained that the system is meant for threat detection and that unmatched biometric data is deleted. Civil liberties organisations dispute this. They argue that continuous surveillance inevitably captures ordinary, innocuous moments—people resting, eating or seeking medical assistance—far beyond any criminal investigation.
Activists have also alleged that protesters were warned that surveillance footage could reach parents, universities or employers. Such warnings, they argue, create a chilling effect by transforming lawful participation into a potentially consequential act. The concern is therefore not simply misuse of information. It is whether the possibility of identification discourages citizens from exercising constitutional rights.
How have other democracies regulated facial recognition?
Countries have adopted sharply different approaches. The European Union's AI Act restricts real-time remote biometric identification in public spaces, allowing only narrowly defined exceptions. Several American cities—including San Francisco, Portland and Boston—have banned or significantly restricted police use of facial recognition. The UK permits live facial recognition but under operational safeguards and judicial oversight. India, however, has no dedicated law governing police deployment of facial recognition.
The Supreme Court's landmark Justice K S Puttaswamy (2017) judgment recognised privacy as an inalienable right under Article 21 and held that restrictions must satisfy the tests of legality, necessity and proportionality.
However, the Digital Personal Data Protection Act, 2023 grants broad exemptions to government agencies, leaving police use of biometric surveillance within an uncertain legal framework.
Is this only about privacy?
Perhaps not. The true power of facial recognition does not lie in the camera but in the database. A photo becomes politically significant only when linked to an identity through police records, passports, driving licences, electoral rolls or other repositories. The issue therefore extends beyond informational privacy to the conditions under which citizens participate in democratic life.
A protester standing at Jantar Mantar may not know whether a camera is recording, whether their face has been matched or whether that information will be retrieved months later. The uncertainty itself becomes disciplinary. The deterrent is not a known punishment. It is the possibility of being permanently recorded.
Traditional policing begins with a crime and searches backwards for suspects. Continuous facial recognition reverses this sequence: it identifies everyone first, preserves the record and allows authorities to reconstruct relationships and associations later.
What safeguards do experts recommend?
Experts commonly suggest four safeguards:
Judicial authorisation: Live facial recognition should require prior judicial approval except in immediate emergencies
Deletion mandates: Biometric information of individuals not linked to wrongdoing should be deleted within a defined time frame
Independent audits: Regularly test systems for accuracy, false positives and algorithmic bias
Purpose limitation: Data collected during protest surveillance should not be cross-referenced with unrelated databases such as educational or employment records
Facial recognition may assist law enforcement when deployed transparently against specific threats. Without safeguards, however, it risks becoming an instrument of indiscriminate surveillance.
The constitutional question?
Democracies have historically depended on a certain degree of forgetting. A citizen could attend a rally, join a strike or stand in a crowd knowing that the moment would eventually fade. Facial recognition removes that possibility. If every public gathering can leave behind a permanent biometric trace, does the constitutional freedom to assemble still include the freedom to participate anonymously?