Delhi Is Building a 10,000-Camera AI Surveillance Network With Sub-Second Facial Matching

TL;DR: Delhi Police is rolling out a permanent, citywide AI surveillance network under its 798-crore-rupee Safe City Project. The system connects 10,000 new facial recognition cameras to a centralized C4I command center built by India's government tech agency C-DAC. It can match 1 million faces in 200 milliseconds. A newly created Picture Intelligence Unit will tap into telecom, banking, and traffic databases. India has no facial recognition law, no biometric data protection statute, and no oversight mechanism for any of this.

From Republic Day Stunt to Permanent Infrastructure

Three days ago, Delhi Police made headlines by deploying AI-powered smart glasses at Republic Day celebrations. Officers wearing AjnaLens devices scanned parade crowds against a database of 65,000 known criminals.^[1]

That was the showcase. Here's the real story.

Delhi Police is building a permanent surveillance network that will blanket India's capital with 10,000 new CCTV cameras. Every one of them compatible with facial recognition software. These cameras join roughly 25,000 already installed across the city.^[2] That's 35,000 cameras feeding into a single, AI-powered nerve center.

The system isn't for special events. It's for everyday policing. And it's being built with no public debate, no legislative authorization, and no privacy safeguards.

Inside the C4I: Delhi's Surveillance Nerve Center

At the core of the project sits the Integrated Command, Control, Communication & Computer Centre, called C4I. Built by the Centre for Development of Advanced Computing (C-DAC), a government research agency, the C4I is designed to be the brain of Delhi's AI policing operations.^[3]

The numbers are staggering:

10,000 new high-resolution cameras with facial recognition capability
Up to 1,000 live feeds monitored simultaneously from the command center
100+ camera feeds analyzed by AI at once
1 million records matched in 200 milliseconds
350,000 known criminals in the facial recognition database
300,000 unidentified body records also stored in the system

The AI can identify over 20 faces in a crowd, even when subjects have partial visibility or are wearing disguises. Officers say the system works through bad lighting, camera blur, and even images that are 10 to 20 years old.^[4]

It doesn't stop at faces. The C4I also integrates automatic number plate recognition, gunshot detection, crowd estimation algorithms, and automatic alerts when someone collapses or appears in distress.

The Picture Intelligence Unit: Where Surveillance Meets Data Mining

Here's where it gets worse.

Delhi Police has created a new division called the Picture Intelligence Unit (PIU). The PIU doesn't just manage camera feeds. It connects facial recognition data to multiple national databases:^[3]

Traffic e-Challan records: linking faces to vehicles
Telecom data: phone records and location information
Banking institution data: financial records

The PIU will also "tag images collected from raids, newspapers, and public submissions to continuously refine the AI's identification accuracy." That means police are crowdsourcing training data from news photos, citizen tips, and their own operations to make the system smarter over time.

This isn't facial recognition. It's a full-spectrum surveillance architecture that combines your face, your car, your phone, and your bank account into a single queryable profile.

Already Making Arrests

The system isn't theoretical. Pilot deployments in North and North West Delhi have already produced results. Since April 2024, police report over 200 arrests in North West Delhi and similar numbers in North Delhi using facial recognition. Between September and November 2024 alone, 70 people were arrested in North Delhi for theft and snatching based on AI-matched identifications.^[2]

Delhi Police frames this as proof the technology works. Privacy advocates see it differently: hundreds of arrests made using a biometric surveillance system that has never been authorized by parliament, reviewed by courts, or tested for racial and demographic bias.

Zero Legal Framework

"Facial recognition is uniquely intrusive: real-time, automated identification at scale, erasing public anonymity," said Apar Gupta, executive director of the Internet Freedom Foundation (IFF). He called the C4I system "a quantum leap in the state's ability to identify and monitor individuals."^[5]

India's Supreme Court recognized privacy as a fundamental right in 2017. But there's no data protection statute that limits how police collect, store, or use biometric data. The government's Digital Personal Data Protection Act of 2023 specifically exempts national security and law enforcement uses, a loophole wide enough to drive a surveillance van through.

No law in India mentions facial recognition. No regulation governs how long biometric data is stored, who can access it, or what happens when the system gets it wrong. Parliament has never debated or authorized this technology.

Compare that to the EU, where the AI Act restricts real-time biometric surveillance in public spaces. Or to San Francisco, Portland, and other U.S. cities that have banned police use of facial recognition entirely. Or even to the UK, where the Met Police's facial recognition program is currently being challenged in the High Court.

India isn't just lagging; it's building one of the world's most expansive urban surveillance systems without any guardrails at all.

The Bias Problem Nobody Wants to Talk About

Even Delhi Police officials acknowledge the system has problems. Demographic biases in AI models, poor camera angles, and adverse weather all affect accuracy.^[5]

Facial recognition systems consistently perform worse on darker-skinned faces, women, and younger people, categories that describe a massive portion of Delhi's population. A 2019 U.S. National Institute of Standards and Technology (NIST) study found that many facial recognition algorithms had error rates 10 to 100 times higher for Black and Asian faces compared to white faces.

When your system matches 1 million records in 200 milliseconds, even a tiny error rate generates a lot of false positives. And in a country where police accountability is minimal and wrongful detention carries few consequences, those false matches don't land equally.

The Bigger Picture: National Surveillance Infrastructure

Delhi's Safe City Project isn't happening in isolation. India's proposed National Automated Facial Recognition System (NAFRS) aims to connect state and central databases into a unified surveillance network covering 1.4 billion people.^[1]

Hyderabad already runs its own extensive facial recognition network. Mumbai, Chennai, and Bengaluru have similar projects in various stages. The Republic Day smart glasses were a visible proof of concept. The C4I is the infrastructure that makes permanent, citywide surveillance possible.

At 798 crore rupees (roughly $94 million), the Safe City Project represents a massive investment in surveillance hardware that, once installed, creates its own momentum. Cameras don't get uninstalled. Databases don't get deleted. Command centers don't get shut down.

What You Can Do

If you live in Delhi: Be aware that facial recognition cameras are being deployed across the city. Public spaces are being surveilled.
Check if you're in the database: India has no mechanism for citizens to verify if their biometric data is stored in police databases or request its deletion.
Support digital rights organizations: The Internet Freedom Foundation and Software Freedom Law Center India are the leading organizations fighting facial recognition overreach in India.
Follow the legal challenges: While India hasn't seen a major FR lawsuit yet, legal challenges in the UK and EU are establishing precedents that could influence Indian courts.

Sources

Published: January 28, 2026