You can’t always tell if someone is recording you with their smart glasses, but your phone can help.

At least **25 people** whom Delhi Police’s **facial recognition system**(FRS) flagged as being present at the Jantar Mantar protests between July 20 and 26 were actually in jail at the time, according to an investigation by The Indian Express. The finding raises questions about the accuracy and reliability of facial recognition systems used by police to monitor public protests. Delhi Police had told the Supreme Court that its facial recognition system identified **2,873 people with criminal records** at the protest. The Indian Express **examined 205 people** on this list who had links to serious offences and found that Delhi prisons held at least 25 of them when the system purportedly detected them at Jantar Mantar. The 25 included 17 people facing murder charges, four facing rape or other sexual-offence charges, including two cases under the POCSO Act, and four facing charges of attempted murder. Three were flagged on July 24, 21 were flagged on July 25, and one on July 26. Police, jail and court records showed that they were in custody when the FRS identified them at the protest. ### **2,873 people flagged by Delhi Police** Delhi Police had told the Supreme Court that its FRS identified 2,873 people with criminal records at Jantar Mantar between July 20 and 26. **Its August 17 affidavit said 2,402 matches** came from the police biometric database, **Crime Kundli** , **while 471 came through criminal records**. The police said further action would follow only after field verification. The finding comes after facial-recognition vans were seen at the protests. MediaNama reported on July 24 that police were using CCTV footage and facial recognition at the site. Videos showed cameras capturing protesters and screens displaying faces inside surveillance vehicles. Delhi Police later told the Supreme Court that they had used the technology only to identify people already in police records, not to surveil the entire crowd. Solicitor General Tushar Mehta said, “**It does not identify everybody indiscriminately**. It captures a face and checks whether it matches the records maintained by the National Crime Records Bureau concerning persons with serious criminal antecedents.” ### **Questions over how facial recognition works** That claim has been questioned on technical grounds. Rohini Lakshané, an independent technologist and researcher, said in the past, “**A live facial-recognition system cannot know who is on a watchlist before performing the comparison**.” She said the system first detects faces and creates biometric templates before checking them against a database. The police also have not disclosed which specific facial-image database they used. The affidavit refers to NCRB records but does not identify the datasets that populated the system. Lakshané said the **distinction matters because being present in a police database does not necessarily mean a court has convicted** the person. The accuracy of the system is another concern. Delhi Police said in a 2022 RTI response that it treated an **80% match as a “positive” result**. But an **80% similarity score does not mean an identification is 80% accurate**. Factors including lighting, camera angle, image quality, masks, the algorithm and the database can affect results. Delhi Police has not published the false-positive rate at its threshold. ### **Reliability of FRS matches under scrutiny** The Indian Express findings therefore raise a basic reliability question: **if the system identified people as present at the protest when records show they were in jail, what level of verification must investigators require before they can treat an FRS match as reliable?** The Supreme Court has allowed the government to proceed against the 2,873 people identified by Delhi Police, while the police maintain that no action will be based solely on an FRS match. The court is also considering the broader legal questions around the technology, including whether its use at the protests was lawful and proportionate. Read more: * What are the risks and challenges posed by facial recognition technology: NITI Aayog * CISF wants facial recognition cameras at airports linked to NATGRID **For You** * Read Reasoned by Nikhil Pahwa: How AI is changing our world * Sign up for MediaNama's Daily Newsletter to receive regular updates * Sponsor a MediaNama Event

A light-hearted take on gendered salutations, identity, and everyday sexism through the recurring phrase ‘Hello madam.’

AI adoption is pressing on the fault lines the media industry has historically failed to address. Among them are rampant tech-facilitated gender-based violence, the deepening of epistemic exclusion and injustice, and the persistent underpayment and devaluation of women’s labour. Article published in 'Pen Power: Media at the Crossroads,' a souvenir edition released at the National Meet of the Network of Women in Media, India (NWMI) held in Thiruvananthapuram, India from 6 to 8 February 2026. This Zenodo record contains two documents: Contains the front and back covers of the souvenir edition, the page with credits to the editoral team and the event's core organising team, and the full text of the article. Contains the full text of the article only. This is better readable on screenreaders and more neurodiversity-friendly.