Five days after the Cockroach Janta Party ended its seven-week sit-in at Jantar Mantar with the resignation of Union Education Minister Dharmendra Pradhan, Delhi Police floated a ₹15.66 crore tender for 338 CCTV cameras fitted with facial recognition. The timing, the siting and the technical specifications of that tender - reviewed by the Internet Freedom Foundation - raise a question that the procurement documents themselves never answer: who, exactly, is meant to be watched, and under what legal authority.
A tender that tracks the protest map
The bid, placed on the Government e-Marketplace on 30 July 2026 and closed on 20 August, covers Central and New Delhi districts under the official justification of securing "VIP/VVIP routes" for Republic Day and Independence Day. But the site list reads like a map of recent dissent. It includes the VVIP gate of Ramlila Maidan, Turkman Gate, Delhi Gate, Rajghat, the India Gate hexagon, Mandi House and Patiala House. On 11 October, the Commissioner of Police de-notified Jantar Mantar as a protest site and designated Ramlila Maidan, along with Burari Ground and CBD Ground Shahdara, as replacements. Of the three, only Ramlila Maidan - and the arteries leading to it - falls inside the camera network already being procured. Burari and Shahdara, both well outside the tender's geography, were left uncovered. There is no evidence the two decisions were coordinated, but the overlap is difficult to ignore.
What the specifications actually demand
The 76-page request for proposal - still labelled "DRAFT RFP 338 CCTV Final" - seeks 248 fixed cameras, 43 pan-tilt-zoom units and 47 automatic number-plate readers, backed by video-management and facial-recognition software with a five-year operating contract. Every fixed and PTZ camera must carry face-recognition analytics alongside "people gathering" detection, a capability whose purpose at sites like Delhi Gate and Ramlila Maidan is difficult to separate from crowd and protest monitoring. Every camera category, including the number-plate readers, comes fitted with a built-in microphone, and the software allows operators to "listen to audio associated with video." Nothing in the tender authorises the recording of sound in a public space, nor does it specify a retention period for voice data, face templates or plate reads. The only storage rule given is a minimum 30 days of video - a disk specification, not a privacy safeguard.
Missing legal scaffolding
A lawful facial-recognition procurement would ordinarily reference the legal order authorising it, a published standard operating procedure, the Supreme Court's privacy jurisprudence in Puttaswamy, and the Digital Personal Data Protection Act. This tender references none of them. There is no privacy impact assessment, no named database against which faces will be matched, no accuracy or similarity threshold, and no mechanism for correcting a false match. That absence matters because Delhi Police has already told the Supreme Court it ran facial recognition on Jantar Mantar crowds in July 2026, generating 2,873 "matches" that subsequent reporting found unreliable. The Ministry of Home Affairs has pointed Parliament to a Standing Order that, when published, made no mention of facial recognition at all. A Supreme Court committee is currently examining the legality of such surveillance, even as the underlying infrastructure continues to expand, district by district, through contracts that bypass the ₹800 crore Safe City programme's own oversight.
A tender riddled with errors
Beyond the legal gaps, the paperwork itself is unreliable. The technical annexure appears copied from an earlier 194-camera tender for North District, complete with that document's page numbers and title. The bill of quantities transposes the number-plate and PTZ camera counts. The video-management software specification is lifted near-verbatim from a single vendor's brochure - a practice that, under the marketplace's own rules, should void the bid. One camera's coordinates place it roughly 40 kilometres east of Delhi, and the integrity pact names anti-corruption monitors whose terms expired years ago, with no reference to the monitor appointed in 2025. In total, 28 such defects have been identified. A system meant to decide, without any declared accuracy threshold, whether a face captured at Delhi Gate belongs on a watch list is being built on documents that cannot even count their own cameras correctly.
