By — Amritesh Unny
Abstract
Between 20th June to 25th July 2026, Delhi Police used real-time Facial Recognition Technology (FRT) on thousands of peaceful student protestors at Jantar Mantar. Indian criminal procedure attaches every safeguard, notice, ground, cognizance, and right, to a protestor who has already been labelled an accused, or a named suspect. One-to-many facial scans reverse that sequence. A machine forms suspicion first, tests it against a criminal database, and crosses no legal threshold to do it. The person being scanned is never informed. That is the void this deployment occupied, and it is the void the courts must now name.
Vision Van
Outside Kerala House, sat a white van bearing the Delhi Police crest, named Ikshana (Sanskrit for vision). Eight cameras on a telescopic mast surveilled the crowd at 360 degrees, a system Delhi Police first inducted during the 2023 G20 Summit. Inside, software drew green boxes around faces and put a number beside each individual it monitored. A Reuters reporter walking past saw his own face boxed on a monitor, with a “face accuracy” match metric beside it. Following the episode, an officer shut the system temporarily, not the surveillance.
Officers wore AjnaLens smart glasses made by Dimension NXG. A fingerprint app, NCRB-Abhigyan, tied to the national fingerprint database, was also being utilized. Drones, body cameras, a mobile command and control vehicle, were all being used with utmost synergy, and the cause a NEET paper leak, a demand that the education minister resign and the families of aggrieved students who committed suicide be given adequate compensation.Uncommonly, the education minister, Dharmendra Pradhan, did resign on July 25th, 2026. What was not ordinary was the apparatus training on the people who sought this resignation.
When law permits the State to measure a citizen
India has a statute for taking a person’s body apart into data. The Criminal Procedure (Identification) Act, 2022, hereinafter referred to as the ‘Act’, defines “measurements” to include finger-impressions, palm and footprints, photographs, iris and retina scans samples, and behavioural attributes. Section 3 of the act states who may be measured, including convicts, persons ordered to give security for good behavior, those arrested for serious offences, and detainees under preventive-detention laws. The National Crime Records Bureau (NCRB), u/s 4(2) of the Act, stores records for 75 years. Refuse, and Section 6 makes your refusal an offence under Section 186 of the Indian Penal Code.
Read the list of people again. It includes, convict, arrestee, and detainee. Every category presupposes that some measure of proceeding has been initiated against you per procedure established by law. The protesters at Jantar Mantar were none of these. Nothing in the Act authorizes the biometric capture of a free citizen standing in a public square. Photographs taken to run facial matches against a criminal database, on people who have not been arrested, squarely fall outside the scheme entirely.
The whole edifice rests on an assumption. In State of Bombay v. Kathi Kalu Oghad & Ors., 1961, an eleven-judge bench held that taking an accused’s fingerprints or handwriting does not violate the bar on self-incrimination apropos Article 20(3), because such marks are identificatory, not testimonial. A foundational permission but noteworthy that the law makes this ruling for an accused. Rationally, if one were to take away the premise, i.e., the rule being for an accused, the permissibility would have nothing to stand on.
Surveillance historically has needed a name
For sixty years, courts allowed the State to watch citizens only when the state could articulate the ‘why’ question, and only against people they could name. In Kharak Singh v. State of U.P., 1963, the Court struck down domiciliary night visits on a history-sheeter. Justice Subba Rao, dissenting, warned that constant watching is a psychological restraint on a free man, a view the Court later adopted. In Gobind v. State of M.P., 1975, surveillance was upheld only against a narrow class, including persons with criminal antecedents, reasonable material, and a compelling interest. In Malak Singh v. State of Punjab & Haryana, (1981), the Court held that police have no license to enter whoever they like in a surveillance register, and that surveillance must not squeeze citizens’ fundamental freedoms or offend their dignity. Even more, in PUCL v. Union of India, 1997, the Court would not let phones be tapped without procedural safeguards they wrote themself.
The commonality that runs through all these caselaw alike is that surveillance began with a specific individual, a recorded reason, and a register a judge could read later when required. History had utilized the limiting principle, and now with facial recognition at a student’s protest completely dissolves years of court-built jurisprudence and induces a horrifying chilling effect in society. The crowd is now on the register. Every face is a probe image checked against the database of the guilty. The limiting principle is gone, a distant dream, and nothing meritorious has replaced it.
The machine as accuser
Here is the inversion, plainly in the old model, an officer formed a suspicion, and the law then tested said suspicion under the given rules. Real-time facial recognition is completely the opposite. The suspicion is generated, beforehand, by an algorithm. Way before any officer decides anything, before any threshold exists, on a person who will never know the wrath they are about to face. Suspicion has been automated and pushed to the front of the sequence, where no protection unfortunately reaches it.
The statistics make it worse, Delhi Police treats a similarity score above 80% as a positive match, i.e., below-80% results are “false positives” that still invite “corroborative” investigation, so a weak match keeps an individual still caught in the net. The force told the Delhi High Court in 2018, in Sadhan Haldar v. NCT of Delhi, 2019, that its own system was “not good” with “only two per cent match”, which by 2019 accuracy had fallen by nearly1%, unable to tell boys from girls. When the ACLU ran Amazon’s tool at that same 80% threshold against 25,000 arrest photos, it falsely matched 28 members of the US Congress to mugshots, with nearly 40% of the errors being people of colour. NIST’s 2019 demographic study found false-positive rates for Asians and African American faces running 10 to 100 times higher than for Caucasians in one-to-one matching, and highest of all for African American women in most searches, which, the report warned, could mean false accusations.
Rajya Sabha MP A.A. Rahim’s Article 32 petition, Diary No. 45049 of 2026, asks the Supreme Court to declare suspicion less biometric surveillance of assemblies, unconstitutional until Parliament legislates, invoking the triple test of legality, legitimate aim, and proportionality, laid down in K.S. Puttaswamy v. Union Of India, 2017. Parallelly, Delhi High Court matter, before Chief Justice D.K. Upadhyaya and Justice Tejas Karia, returns on 25 August 2026, with the bench having instructed the petitioner to re-examine the existing framework and published SOPs first.
Silver Bullet: The piercing kind
Concludingly, there is a tell in the law itself that a private member’s bill, the Facial Recognition Technology (Regulation of Police Powers) Bill, 2023, has sat in the Rajya Sabha since December 2023. It expressly treats FRT as distinct from “measurements” under the pre-existing 2022 Act, and would require a magistrate’s order before deployment, with an express right to be heard, and deletion on acquittal or discharge. Parliament’s own draft concedes to the same point, that the 2022 Act, which is the current bulwark against oppression, does not hold any mechanism against tyrannical use of FRT. The scanning at Jantar Mantar was not authorised de jure. It was done in a legal vacuum. The question in the title has a precise answer today, which is that being scanned makes you a suspect the moment a machine decides it does. Which goes to show that today you can be herald a suspect by an artificial being, long before that law ever recognizes you as so.
About the Author:
Amritesh Unny is a three-year LL.B. (Hons.) student at Jindal Global Law School, O.P. Jindal Global University. He has a deep interest in human rights.
Image Source: Jagran.com