Five days after the Cockroach Janta Party called off its weeks-long sit-in at Jantar Mantar in July 2026, Delhi Police floated a ₹15.66 crore tender for 338 cameras fitted with facial recognition, people-counting analytics and built-in microphones. The sites named in the bid documents trace the roads connecting Jantar Mantar, Parliament, Ramlila Maidan and India Gate - and when the force later shifted its designated protest ground to Ramlila Maidan in October, it moved the demonstrations onto territory it had already marked for surveillance.
A Tender Built Around Watching Crowds
The request for proposal, reviewed by the Internet Freedom Foundation, asks for 248 fixed cameras, 43 pan-tilt-zoom units and 47 automatic number-plate readers, each required to run facial recognition, "people gathering" detection and audio capture. The specifications are explicit: cameras must be positioned "to identify and capture the face of a particular person roaming around at Critical/vulnerable points." This is not a passive security camera network. It is an identification system, designed from the outset to flag individuals in a crowd rather than simply record a scene.
The site list reads like a map of protest geography. It covers Ramlila Maidan's VVIP gate, Delhi Gate, Turkman Gate, Rajghat, the India Gate hexagon, Mandi House and the approaches to Patiala House courts. Two of the three newly designated protest grounds, Burari and Shahdara, sit outside the tender's reach. Ramlila Maidan, barely three kilometres from Jantar Mantar, sits squarely inside it.
Legal Gaps Where Safeguards Should Be
Facial recognition technology works by converting an image of a face into a mathematical template and comparing it against a reference database, flagging matches above a set similarity threshold. None of that architecture is specified here. The tender does not say which database faces will be checked against, what threshold counts as a match, who is authorised to run a search, or how a wrongful identification gets corrected. There is no privacy impact assessment and no citation of the Digital Personal Data Protection Act or the Supreme Court's privacy ruling in *Puttaswamy*. The only retention figure given, a minimum 30 days, describes disk capacity rather than a data-handling policy - it says nothing about how long face templates or audio recordings are kept once captured.
This matters because Delhi Police has already told the Supreme Court it ran facial recognition on the Jantar Mantar crowds, generating 2,873 claimed "matches" that reporting has since called unreliable. The surveillance operated under a Standing Order the Home Ministry has declined to make public and which, once disclosed, turned out not to mention facial recognition at all. A Supreme Court committee is currently examining the legality of such protest surveillance, even as procurement for more of it proceeds on a separate track.
A Pattern Repeating District by District
This is not an isolated purchase. The same Provisioning & Logistics unit had already installed 6,630 cameras across fifty police-station areas, roughly one in ten carrying face recognition, according to an earlier RTI disclosure. An identical template produced a 194-camera tender for North District months earlier, and the specification document attached to this new bid still carries that tender's title and page numbers - evidence of documents copied wholesale rather than drafted for purpose. The bill of quantities confuses pan-tilt-zoom cameras with number-plate readers, the video management software specification lifts language straight from a single vendor's product brochure, and the integrity-pact monitors named to guard against corruption left that role years ago under vigilance rules.
None of this proves intent to target dissent specifically. What it shows is a procurement process moving fast, with little regard for internal consistency, toward a system with no stated accuracy threshold, no retention law and no public accountability - on streets the state has also chosen as the designated place to protest.