Daybreak - Delhi built a ‘safe’ surveillance city. It can now locate anyone in seconds

Episode Date: October 5, 2026

Your face caught on a camera, your vehicle number plate, a voter ID, a driving licence, even a police-clearance certificate—any of these can become part of a vast web of data that the police can search in Delhi.That matters especially now, as the capital sees mass protests and public demonstrations, raising questions around safety. surveillance, privacy and the right to dissent.The Ken's Pratap Vikram Singh spent a month investigating Delhi’s ₹857-crore Safe City Project, conceived in the name of women’s safety. Its mandate now extends far beyond that, to crime prevention, public-order surveillance and political and religious rallies.What exactly is Delhi building, who can access it, and what happens when a system designed to find criminals can find almost anyone? In this episode, host Snigdha Sharma reads Pratap’s report titled: Delhi’s mass surveillance: track 50 million faces, identify them in five secondsDaybreak is produced from the newsroom of The Ken, India’s first subscriber-only business news platform. Subscribe for more exclusive, deeply-reported, and analytical business stories.

your face caught on a camera, your vehicle number plate, voter ID, driving license, even a police clearance certificate. Any of these can now become a part of a vast web of data that the police can search in Delhi. And that matters especially now when the national capital is seeing mass protests and public demonstrations. And questions around surveillance, privacy and the right to dissent are becoming harder to ignore. My colleague, the Ken reporter Pratap Vikram Singh, spent one month investigating one of the most ambitious surveillance systems being built in the national capital. The Delhi Police's 857 crore rupees Safe City Project. And here this. It was conceived in the name of women's safety.

That irony feels particularly stark now, as Delhi sees protests over, among other things, the lack of safety for women in the national capital. But the mandate of this project now extends far beyond that to crime prevention, public order surveillance and political and religious rallies. So what exactly has Delhi built? And what happens when a system designed to find criminals can almost also find anyone else? That is the investigation that I am going to read out aloud to you today in Pratab's own words. Welcome to Daybreak, a business podcast from the Ken. I'm your host, I'm Dhan Sharma and I don't chase the news cycle. Instead, every day of the week, my colleague Rachel Vargheese and I

will come to you with one business story that's worth understanding and worth your time. Today is Tuesday the 6th of October. If you want to go around Delhi and expect not to be pursued by prying eyes, good luck. If any of the mischrooming crop of cameras installed around the capital catches a glimpse of your face or vehicle number plate, law enforcement officials can identify you and track you down. without breaking a sweat. Especially if you have a voter's ID or a driver's license or a vehicle registration certificate

or if you're a tenant or a domestic worker or if you or your employer has ever applied for or obtained a police clearance certificate or a character certificate or if you have been named in a court case anywhere in India. You have landed in the grid and there is no escaping it. Consider this. In April, a 37-year-old man driving a white Tata Sierra plowed through a security barricade at the entrance to Delhi's Legislative Assembly, sparking a security scare. He stepped out, placed what appeared to be a bouquet in the speaker's parked car and fled. The man was caught in less than two hours.

The Sierra was stopped at a police picket barely four-kilometer from the assembly and the driver detained. The police did not get lucky. They had help. City project. That's a benign sounding name for an 857 crore-ru-Rupe AI-based mass surveillance system that the Delhi police have been quietly putting in place over the last four years. It searches faces against a web of databases. Police clearance certificate, interoperable criminal justice system, electoral photo identity card, wahan, Sarathie to identify a person of interest. It is a data lake of pictures, videos and personal and personal and

biometric information of nearly 4 million individuals and counting. To access these, technology providers like NEC and Awee Ros have built applications and search engines for the Delhi Police. While the Japan headquartered NEC is the system integrator, Gurgao-based Awee Rose provides core applications like AI video analytics, facial recognition and a reverse image surge engine called Anvation. The Ken conducted a month-long investigation, interviewing at least 20 police officers, service providers and consultants

and reviewing official documents related to the Safe City project to unravel the workings of one of the most pervasive surveillance systems deployed anywhere in the country. The first phase of the federally funded project was rolled out in February and the plan is to have it fully operational by the end of the year, at which point, at least one person at each of the city's 200-odd police stations will carry a smartphone with a facial recognition app. It could simply point the phone at a person of interest, click a picture and almost instantly pull up their details.

And anyone could be a person of interest. The Safe City project began in the name of women's safety. It grew out of a plan devised in the backdrop of the 2012 Delhi gang rape and paid for by the central government's Nirbaya Fund to help prevent heinous crimes by boosting the video surveillance capabilities of the police. The idea was to blanket the city with surveillance cameras and use AI to analyze the footage to identify women in distress and rush aid to them, or gather intelligence and evidence for the police investigations. The system would be fed by the voluminous criminal and citizen

databases held by government agencies. The CDAC, which is the total service provider for this project under METI said in its 2021 request for proposal. A former Delhi police official and a former tech executive who worked on the Safe City project said that it was also designed to use Adhah demographic data. The Ken could not verify the claim and the police officials denied it. Typically, facial recognition systems are designed to rely on criminal databases alone. Privacy rights advocates question the utility of creating a general purpose mass surveillance system and argue that using citizen databases for AI surveillance, is illegal.

Some police officers and service providers say that there is a need for regulation to install safeguards and ensure accountability. It is very easy to track someone to the last mile. A former Vero's executive who was closely involved with the project argued, pointing to the leak of the personal details of some of the people at the quote-unquote cockroach protests in Delhi where the police used the company's facial recognition system. The government should bring in guardrail. to prevent any misuse, he said.

One of the key objectives for Safe City, according to the request for proposal, is to maintain public order by conducting surveillance of large events such as political and religious rallies. Women's safety was just a cover, said a CDAC official who, like most people the Ken spoke with, requested not to be named. The project aims to strengthen security against terrorism and serious crimes, they said. The RFP provides for deploying nearly 10,000 cameras and connecting the existing 15,000 installed by the police under various schemes.

Nearly 3,000 new cameras have also been installed and 5,000 existing cameras integrated into the system so far. Their feeds going to the integrated control and command center at the police headquarters. The hardware and the software for both facial recognition and automated number plate recognition, the latter supplied by Noida-based Wehant technologies, have been tested and are close to But the NEC is still to deliver the capability to analyze feeds from multiple cameras in real time using AI. Several other AI analytics functions like finding distressed women and crowd counting, the system

should be able to identify at least 20 faces in a crowd, are also yet to be fully integrated. Once the data lake is fully expanded and operationalized, the request for proposal or RFP declares law enforcement agencies can use it to do proactive police. and even predict crimes. By analyzing historical crime data, for example, they can determine, and I'm quoting, when certain areas are more vulnerable, or if juvenile offenders can become, quote-unquote, major criminals.

They can identify, and I'm quoting again, vulnerable groups likely to be hurt as religious targets or people who are at risk of domestic violence. Facial recognition works by converting a face, its shape, its size, spacing of eyes, forehead, jaw, mouth, ears into a binary signature and matching that against images in the connected databases. It is the same principle behind the unlock with face feature of the smartphone. You can track a person either through facial appearance or through body posture, said the former

Averos executive. The accuracy is lower in the latter, but at least you are giving the police something to work with. In 2023 to 24, the Delhi Police ran. ran a series of tests to assess the facial recognition and license plate reading functionalities of the system. We fed it images of over 80,000 police personnel and used cameras in and around the headquarters to capture them.

The Averos executive said, explaining the early days of developing the technology. Then, in the first half of 2024, the police tested the facial recognition system on people with criminal histories. In one test, they identified a set of 81 suspects and convicts out on top. parole, gave their photos to Aviros and told them to move in front of cameras installed at select locations, including Ashokvihar, Horskhas, Gita Colony, Mukherjee Nagar and Najafkar. The system identified over half of the criminals, a former Averroos executive said. The firm said in response to queries about the accuracy of the tests sent by the Ken

and I'm quoting, any testing or trials conducted by Delhi police are entirely within their prerogative and the V-Rose cannot comment on or confirm whether any such activity was conducted. This raises the question. Who controls what the system can do if it is entirely in the police's hands? The police are yet to finalize a standard operating procedure for working the system. They are yet to decide who all can ask for video footage and who can approve those requests, a person close to NEC said. What they are settled on is the need to access all more. most limitless data. A former Delhi police chief argued defending the sweeping scope of the project, and I'm quoting, if the police are to do their job, they need tools, whether they are achieved by

connecting 32 databases or 1,000. Look at Delhi east of the Yamuna, he said. People don't know who is residing next door. This was the thinking that informed the NEC when it started creating a data lake for the system in 2024, integrating over two dozen databases, maintaining by the Delhi Police, the Transport Ministry, the Election Commission and the courts. The largest of these databases, the police clearance certificate alone contains the personal details of 2.5 million people. To scour the databases, NEC and Aviro's developed enterprise search and Unvation respectively. Envation works like Google's reverse image search. The police can search by names, Elias, the colour of clothes, vehicles, bags and such like. It even allows for

retrospective search showing all the places where the person had been captured by a camera, body worn, smartphone or drone over the past 30 days, the period of which data is stored. The RFP or the request for proposal envisions facial recognition to happen in real time or offline mode. You can query the system for people of interest, vehicles of interest, places of interest and databases of interest. You can log in and get all the feeds in one place. You can search for a person on a black motorcycle wearing a yellow helmet and carrying a pink or black backpack, the former police official explained. Oftentimes, victims of crime give you such descriptions. There are over 30 fields of information about every person in the system and doing a search

returns the name, phone number, address, blood group, driver's license, passport and Adhar. The system, the government's RFP states, laying out the penopticon design, shall be given access to databases of passport, CCTV, prisons, AMBID or open source databases or any similar database available within the state and central governments. It shall also create a repository of photographs obtained from various police sources like newspapers, photos during raids, photos sent by people, etc. Such photos would be tagged for sex, AIDS, scars, tattoos, etc., so that they become searchable. The project though is running behind schedule. The data lake and the analytics applications are almost ready, but CDAC and NEC and Averos are yet to deliver a fully functional system.

They are now supposed to do so by the end of 26. A Veros was flagged by the Bangalore Police for delivering poor accuracy in August 2025. A letter sent by the police to its system integrator, Honeywell and reviewed by the Ken, sought to replace Averos for video analytics and facial recognition. The company has since claimed to have improved its work and the police have apparently withdrawn the letter. Averos supplies facial recognition tech to more than 10 cities and states, including Jammu and Kashmir. Its interest in the Safe City project came early. Three of the seven bidders of the Delhi Safe City proposed Averos's video analytics solution as a part of their respective bids, the company said. The RFP of the proposal prescribed a unified video intelligence platform,

that could support multi-wender applications. But even NEC lacked the features back then. Aviros had it. The Delhi police started deploying CCTV in bulk around 2010 when the city was preparing to host the Commonwealth Games. By the time the CDAC released the RFP, the police was already using facial recognition if sporadically and had just issued a contract for Bharat electronics

to install over 6,000 cameras, nearly 600 of which would read number plates. Some police units also bought facial recognition tools from Delhi-based in Nefu Labs and the Israeli firm Korsight inviting criticism from human rights and civil society groups. Condemnation only grew after the police used facial recognition to charge several people whom they alleged were involved in the protests against the Citizenship Amendment Act in 2020 and the Delhi riots the same year. Such opposition, however, failed to stall the expansion of the surveillance system

which was most recently deployed at the Cockroath-Chanta Party protest. The police claimed that the system identified 2,800 people with criminal histories at the protest site, although they wouldn't clarify if any of them had committed any offence by just being there. In the aftermath, personal details of many protesters, including minors, were leaked and the police personnel and vigilantes landed at their homes demanding apologies. The police denied the leaks were linked to their surveillance system. They also told the Supreme Court in August that they did not use facial recognition

to track protesters, but only people with criminal antecedents. The design and the deployment of the system, though, tell another story. The police say that their database contains images of about 300,000 criminals. The RFP or the proposal, however, requires, city to support a watch list of up to 50 million unique faces, exceeding the population of the National Capital Region, and identify a person of interest within five seconds. The police are making the system available to all their stations in the city as a mobile app.

They have already given the keys to at least 182 stations, a CDAC official said. The RFP or the proposal provided for system administrators to have full access to the data Lake and users limited access. The roughly 200 users, one at each police station, could raise a request for a search, but it would need to be approved by a senior official, a former Averos executive explained. And I'm quoting again, the Delhi police pressed the vendors to remove the approval layer and provide a direct search facility at the user level. A former special commissioner of police argued that the project ran counter to the spirit of

the digital personal data protection. Act. Though the 2023 law was still not fully notified, he said it gave sweeping exemptions to law enforcement agencies, which was a crucial gap that needed fixing. Apar Gupta from the Internet Freedom Foundation and advocacy group said that the law applied to consensual data collection and required the user to be notified. Surveillance, by its very designed, relies on the absence of notice or consent. That is why we need laws to govern surveillance. Gupta argued. We need legislation that can firstly prohibit any kind of mass surveillance. The Supreme Court's ruling on the fundamental right to privacy necessitated and enabling legislation

to define the ambit, purpose and thus limits of surveillance activity, he said. David Nolan, who is a senior investigator researcher at Amnesty International, said, and I'm quoting, International law in India's own Supreme Court in KS Putaswami v. Union of India, hold that any restrictions to the right of privacy must meet the tests of legality, necessity and proportionality. Facial recognition technology that uses indiscriminate biometric data fails all three by design and India's legal framework also does not bridge this gap.

That is why the government kept a loophole in the DPDP Act. Justice BN Sri Krishna, a retired Supreme Court judge who led the government's expert panel on the data protection framework said. He also said that the legislation does not state what should be done to balance constitutional rights against the state's ability to infringe privacy in certain circumstances. The law grants law enforcement agencies blanket exemption on the grounds of national security.

If it is about tracking criminals and terrorists, it is still understandable. But as a citizen, when I apply for a driver's license, I expect that the data would be used, for processing my license, not for tracking me because of my political thoughts. This is wholly illegal. This should be challenged in court, Justice Srikrishna said. What is the point of digital data protection law when you don't have any protection?

What is most concerning is that the Safe City project is not tied to any specific purpose. It varies from crimes against women to law and order and to surveillance of political and religious rallies to preventive policing, Gupta said. It is a perilous system for citizens' rights when its capabilities are expanding, but the purpose and the guardrails remain fuzzy. Daybreak is produced from the newsroom of the Ken, India's first subscriber-focused business news platform. What you're listening to is just a small sample of a subscriber-only offerings and a full subscription unlocks daily long-form feature stories, newsletters and podcasts.

Head to the Ken.com and click on the red subscribe button on the top of the website. Today's episode was hosted and produced by Snigash Irma and edited by Arna Prakash.