Skip to main content
Medium reported severity Investigation Opened

Delhi: the police facial-recognition system logged at least 25 people who were in Tihar, Mandoli or Rohini jails as present at the Jantar Mantar student protests (20 to 26 July 2026), with timestamps on 24 to 26 July, on a sworn list of 2,873 persons with 'criminal antecedents' that the Supreme Court allowed police to register a fresh FIR against (Indian Express investigation, 4 September 2026)

In an affidavit of 17 August 2026 before the Supreme Court of India, Delhi Police said its Facial Recognition System (FRS) had identified 2,873 people with criminal antecedents at the Jantar Mantar protests of 20 to 26 July 2026 (2,402 through its 'Crime Kundli' biometric database and 471 through other criminal records). On 1 September the Court quashed the FIRs against the student protesters but let the police proceed against the 2,873; the police say any action will follow field verification. The Indian Express checked the 205 listed people facing murder, attempted-murder, rape or child-sexual-offence charges against police, prison and court records and found that at least 25 of them (17 accused of murder, four of rape (two or three under POCSO; the Express's narrative and its list differ), four of attempted murder) were lodged in Delhi's Tihar, Mandoli or Rohini prisons when the system logged them at the protest site with timestamps on 24, 25 and 26 July; some had been in custody for years. Delhi Police told the newspaper that further verification of the 2,873 was pending and, in its affidavit, that no action is taken solely on a facial-recognition result and that field verification follows each match. The police disclosed in 2022, in reply to a Right to Information request, that it treats a match with an 80 per cent similarity score as positive. Whether any of the 25 has since been named in the fresh FIR or visited for verification is not reported.

AI system
Unidentified facial recognition system
Occurred
24 Jul 2026 to 26 Jul 2026
Reported
4 September 2026
Event location
India
What the AI did
Made a claim about the person
Reported harm
Legal Harm
Whose AI use
An institution’s AI use
Setting
Justice · Public services · Privacy
Evidence
AI involvement reported · Causal attribution supported · 4 sources, 2 underlying accounts
5 claims: 5 reported. 7 open questions
People reported harmed
At least 25 people

AI system as recorded: Delhi Police Facial Recognition System (FRS): face-matching software run on protest footage against the police 'Crime Kundli' biometric database and other criminal records; the vendor and algorithm are not identified in the inspected reporting

What Happened

The Indian Express (4 September 2026) reported that when the Supreme Court quashed all FIRs relating to the exam-leak protests by students on 1 September, it let the government proceed against the 2,873 people whom Delhi Police claimed had criminal records and had been identified by its facial-recognition software at the main protest site at Jantar Mantar between 20 and 26 July, the police having said that any action would follow field verification. Of the 2,873 people in the police affidavit, the newspaper focused on the 205 accused of murder (101), rape (61), offences under the Protection of Children from Sexual Offences Act (six) and 37 of the 62 listed under attempt to murder, and found from police, jail and court records that at least 25 of them were in jail when the FRS flagged them: 17 facing murder charges, four rape charges (the narrative says two POCSO cases, the itemised list marks three) and four attempted murder, lodged in the Tihar, Mandoli and Rohini prison complexes. The newspaper lists each with the FRS timestamp (three on 24 July, twenty-one on 25 July, one on 26 July) and the custody dates, which in some cases go back years; at least one is an accused in a high-profile Delhi murder. Responding on 31 August to the newspaper's questionnaire, Delhi Police said the 2,873 had been spotted as prima facie having criminal antecedents and that further verification was pending. The affidavit, submitted on 17 August in response to petitions alleging excessive force and surveillance at the protest, said the FRS identified people whose particulars matched official police databases, 2,402 through the 'Crime Kundli' biometric database and 471 through criminal records; it does not say whether those identified were accused, convicted or merely named in cases, states that only people facing serious offences are held in police face records, and says that no action is taken solely on the FRS result, which is the first step before field verification. The newspaper recalled the police's 2022 reply to a Right to Information request that it treats an 80 per cent accuracy rate as a positive match, and noted that the result is not by itself proof of identity (Outlook adds that a match does not prove presence at the protest). Outlook India relayed the findings the same day. The Internet Freedom Foundation, whose 2022 RTI replies disclosed the 80 per cent threshold, published an analysis calling the field-verification process itself the harm (police visits to homes and workplaces communicating to family, landlords and employers that a person is on a protest-linked list) and demanding withdrawal of the list, disclosure to each listed person and a moratorium on facial recognition at assemblies. The Federal noted that the Court had on 18 August constituted a five-member High-Powered Enquiry Committee headed by Justice R Subhash Reddy whose terms include surveillance, that a petition by MP A A Rahim challenging the use of facial recognition at the protest is pending, and that the Court's order made no finding that the 2,873 were present or had qualifying records, and that the permitted FIR is 'with respect to' the listed persons rather than an instruction to arrest them.

Reported harm

At least 25 people in judicial custody were recorded by the police facial-recognition system as present at a protest they could not have attended and were placed, in a sworn affidavit before the Supreme Court, on a list of 2,873 persons with criminal antecedents against whom the Court permitted a fresh FIR and police field verification; The Indian Express reports the misidentifications and the Internet Freedom Foundation describes the exposure to verification visits and prosecution as the harm. No arrest, FIR entry or verification visit of any of the 25 is reported in the inspected sources.

Outcome

Ongoing

Supreme Court order of 1 September 2026 (a Bench headed by the Chief Justice, invoking Article 142) quashed the FIRs against students and permitted Delhi Police to register a fresh, specific FIR concerning the 2,873 persons with alleged serious antecedents, directing that this should not prejudice their rights (IFF). The Federal notes the order prescribed no method or express requirement that verification precede arrest or other action; the field-verification undertaking is the police's own. Delhi Police's affidavit of 17 August 2026 states that no action is taken solely on an FRS result and that field verification is done for each recognised face. Responding to The Indian Express on 31 August, the police said the 2,873 had been spotted as prima facie having criminal antecedents and that further verification was pending; the response did not address the 25 cases. The Internet Freedom Foundation, which sent a legal representation to the Commissioner of Police on 24 July 2026 and filed RTI applications, has called for the list to be withdrawn and the findings placed before the Court. No later action is reported in the inspected sources.

What remains unknown

  • Whether any of the 25 has been named in the fresh FIR, visited for field verification or otherwise acted against.
  • Whether the 25 (or the other 2,848 listed people) have been told they are on the list.
  • How many of the remaining 2,668 entries would fail the same check; the Express examined only the 205 most serious.
  • The vendor, algorithm and threshold actually applied at Jantar Mantar in July 2026.
  • The full text of the 17 August affidavit and the 1 September order (described as quoted).
  • Whether two or three of the four rape accused are POCSO cases: The Indian Express's narrative says two, its itemised list says three.
  • The protest window: The Indian Express and Outlook give 20 to 26 July; The Federal gives 20 to 25 July.

What the evidence supports

AI involvement: reported. Delhi Police's own affidavit, as quoted by The Indian Express, attributes the 2,873 identifications to its Facial Recognition System matching faces against the 'Crime Kundli' biometric database and criminal records; the Express's records check shows the system logged at least 25 people who were in custody. The affidavit was not inspected directly.

5 claims: 5 reported. What the statuses mean

Reported In an affidavit of 17 August 2026 before the Supreme Court, Delhi Police said its Facial Recognition System had identified 2,873 people with criminal antecedents at the Jantar Mantar protest site between 20 and 26 July 2026, 2,402 through its 'Crime Kundli' biometric database and 471 through criminal records, and that no action is taken solely on an FRS result before field verification.

Causal attribution. The affidavit as quoted by The Indian Express and relayed by Outlook and IFF; not inspected directly.

  • indianexpress.com(opens in new tab) supports · English
    'According to the police affidavit submitted on August 17, its FRS identified individuals with criminal records at the protest site'; 'It said that 2,402 people were identified by the Delhi Police'; 'and 471 through criminal records.'; 'No actions are taken solely based upon the result of the facial recognition system which is only the first step to identify a person in the crowd who has a criminal record.'
  • outlookindia.com(opens in new tab) supports · English
    'According to the Delhi Police affidavit, 2,402 people were identified through its biometric database, Crime Kundli, while another 471 were identified through criminal records.'
  • internetfreedom.in(opens in new tab) supports · English
    'The Delhi Police's affidavit of August 17, 2026 told the Supreme Court that 2,873 persons with criminal records had been spotted by the FRS at the protest site.'
Reported At least 25 of the 205 listed people facing murder, rape, POCSO or attempted-murder charges were in Tihar, Mandoli or Rohini jails when the FRS logged them at Jantar Mantar (17 murder, four rape of which the Express's narrative counts two and its itemised list three as POCSO cases, four attempted murder), with timestamps on 24, 25 and 26 July 2026 and custody dates that in some cases precede the protest by years.

Causal attribution. The Indian Express's own check of police, prison and court records (one chain, relayed by Outlook and IFF); the Delhi Police response did not address the 25 cases.

  • indianexpress.com(opens in new tab) supports · English
    'An investigation of those with serious charges, murder, attempted murder and rape, and child abuse, shows that at least 25 people on the list, identified by the Facial Recognition System (FRS) at Jantar Mantar, were in jail as per police, jail and court records.'; 'In this group, 17 faced charges of murder; four of rape, including two in POCSO cases; and four more of attempt to murder were lodged in Delhi'; 'Court and jail records for all 25 show custody dates precede the protest'
  • outlookindia.com(opens in new tab) supports · English
    'At least 25 of the 2,873 people identified by Delhi Police'; 'were already in jail when they were flagged, an investigation of police, prison and court records by The Indian Express has found.'
  • internetfreedom.in(opens in new tab) supports · English
    'Three were "identified" on July 24, twenty-one on July 25 and one on July 26, each with a timestamp.'
Reported On 1 September 2026 the Supreme Court quashed the FIRs against the student protesters but permitted Delhi Police to register a fresh FIR concerning the 2,873 persons with alleged serious antecedents, directing that this should not prejudice their rights, and without a finding that they were present; the undertaking that action would follow field verification is the police's own, and The Federal notes the order prescribed no such requirement.

Causal attribution. The order as described by the three outlets; not inspected directly.

  • indianexpress.com(opens in new tab) supports · English
    'On Tuesday, when the Supreme Court quashed all FIRs related to the exam leak protests by students, it let the government proceed against the 2,873 people who the Delhi Police claimed had criminal records'
  • thefederal.com(opens in new tab) supports · English
    'Delhi Police was given permission to register one fresh FIR concerning 2,873 people alleged to have serious criminal antecedents and initially reported to have been present at Jantar Mantar.'; 'The Court did not find that the 2,873 were present, had qualifying criminal records or committed violence.'; 'the order prescribed no method or express requirement that checks precede arrest or other action'
  • internetfreedom.in(opens in new tab) supports · English
    'On September 1, 2026, a Bench headed by the Chief Justice of India, invoking Article 142, quashed all FIRs registered against students who participated in the July protests'; 'while directing that this should not prejudice their rights and that they must be given every opportunity to defend themselves'
Reported Responding to The Indian Express, Delhi Police said the 2,873 individuals had been spotted as prima facie having criminal antecedents and that further verification was pending; the response did not address the 25 cases.

Causal attribution. Police statement as quoted by the Express.

  • indianexpress.com(opens in new tab) supports · English
    'Responding on Monday to a questionnaire from the Express, the Delhi Police said:'; 'During the protests, 2873 individuals identified as prima facie having criminal antecedents were spotted. However, further verification in this regard is pending.'
  • internetfreedom.in(opens in new tab) supports · English
    'In its response to the Express, the Delhi Police did not address the 25 cases'
Reported Delhi Police disclosed in 2022, in reply to Right to Information requests, that it treats a facial-recognition match with an 80 per cent similarity or accuracy score as positive.

Causal attribution. Delhi Police's 2022 RTI replies as described by IFF, which obtained them, and by The Indian Express, which IFF says relies on the same disclosures; one chain. The reply itself was not inspected. IE describes the threshold as an accuracy rate of 80 per cent and IFF as a similarity score above 80 per cent.

  • indianexpress.com(opens in new tab) supports · English
    'In response to a Right to Information (RTI) request in 2022, the police had revealed that it considered a match to be'; 'if its system showed an accuracy rate of 80 per cent.'
  • internetfreedom.in(opens in new tab) supports · English
    'the Delhi Police disclosed that it treats any match with a similarity score above 80% as "positive"'; 'extracted in 2022 only after a Second Appeal before the Central Information Commission'

Sources

4 sources inspected, from 2 underlying accounts. Sources that repeat one account do not corroborate each other.

How the sources were read, and where the events happened

Read on 2026-09-24 from an Internet Archive capture of 15 September 2026 (web.archive.org/web/20260915075622id_/...); the live page returns 403 to this host. Express Investigation of 4 September 2026 with the list of the 25 and the police response. Applies to s1.

Read live on 2026-09-24 (Outlook India, 4 September 2026): relays the Express findings and the affidavit figures; same chain. Applies to s2.

Read live on 2026-09-24 (Internet Freedom Foundation, Apar Gupta, 4 September 2026): advocacy analysis that relays the Express findings (same chain for the 25) and adds IFF's own 2022 RTI replies on the 80 per cent threshold, its 24 July 2026 representation and RTI applications. Applies to s3.

Read live on 2026-09-24 (The Federal, 4 September 2026): analysis of the Supreme Court order and the affidavit relying on the Express report; same chain. Applies to s4.

Event countries: India. Affected-person countries: India. Court countries: India.

The protests and the facial-recognition deployment were at Jantar Mantar, New Delhi; the 25 were in Delhi's Tihar, Mandoli and Rohini prisons; the affidavit and order are in the Supreme Court of India (The Indian Express, IFF, The Federal).

Reviewed for publication 2026-09-24: Published under the 2026-09-15 charter as a contextual case in which an AI system made a consequential claim about identifiable people: a police facial-recognition system placed at least 25 people who were in prison at a protest, on a sworn list that the Supreme Court permitted police to act on. The finding rests on The Indian Express's records investigation (one chain, relayed by Outlook, IFF and The Federal) with the affidavit and order described as quoted; the 80 per cent threshold rests on IFF's 2022 RTI replies, which IFF says the Express also relies on. The listed people are not named.

People described

At least 25 people, all in judicial custody in Delhi prisons at the time, whom the police facial-recognition system logged as present at the Jantar Mantar protests and who appear on the affidavit's list of 2,873 persons with criminal antecedents; they face murder, rape, POCSO or attempted-murder charges in unrelated cases. Named by The Indian Express; not named here

People reported harmed in this case

At least 25 people

0 AI participants · 25 other people harmed

The Indian Express states 'at least 25 people on the list' were in jail when logged, and lists them (17 murder, four rape, four attempted murder). The other 2,668 entries were not checked, so the count is a documented minimum of 25.

Counted once within this case. The same person may appear in other cases. This count does not establish AI causation.

Cite this case

Compiled per our published methodology: verification statuses, sourcing standards, and corrections process.

APA

NOPE. (2026). Delhi: the police facial-recognition system logged at least 25 people who were in Tihar, Mandoli or Rohini jails as present at the Jantar Mantar student protests (20 to 26 July 2026), with timestamps on 24 to 26 July, on a sworn list of 2,873 persons with 'criminal antecedents' that the Supreme Court allowed police to register a fresh FIR against (Indian Express investigation, 4 September 2026). AI incidents. https://nope.net/incidents/2026-delhi-police-facial-recognition-jantar-mantar-list-25-in-jail

BibTeX

@misc{2026_delhi_police_facial_recognition_jantar_mantar_list_25_in_jail,
  title = {Delhi: the police facial-recognition system logged at least 25 people who were in Tihar, Mandoli or Rohini jails as present at the Jantar Mantar student protests (20 to 26 July 2026), with timestamps on 24 to 26 July, on a sworn list of 2,873 persons with 'criminal antecedents' that the Supreme Court allowed police to register a fresh FIR against (Indian Express investigation, 4 September 2026)},
  author = {NOPE},
  year = {2026},
  howpublished = {AI incidents},
  url = {https://nope.net/incidents/2026-delhi-police-facial-recognition-jantar-mantar-list-25-in-jail}
}

Related cases

High Unidentified facial recognition system

Jacksonville, Florida: a 38-year-old father was arrested at work in May 2025 and charged with soliciting a child after sheriff's investigators identified him, he says with facial-recognition tools, from photos an impostor had sent to an undercover officer; charges were dropped in August 2025 after the real sender confessed, and the family says the ordeal cost $45,000

The Florida Trib (28 September 2026, republished by the Florida Phoenix) reports that in May 2025, during the multi-agency sting Operation Rescue Our Children, a man using the Skout dating app sent an undercover officer posing as a 13-year-old girl two photos of someone else: a former friend, a supermarket assistant manager and father of two whom the Trib calls 'Nick'. Investigators told him they relied on facial-recognition tools to identify him in the pictures, he recalled, and the Trib found no indication in court documents of other investigative work before his arrest. He was handcuffed at work, charged with soliciting a child via computer and unlawful use of a two-way communications device, spent a night in jail, wore an ankle monitor, was barred from contact with his two young children for a month and a half and could not live at home for nearly three months. The account's owner, a man with a prior conviction for the same offence, admitted on 31 July 2025 using the photos; charges against the father were dropped the next day and a judge ordered the case expunged on 4 August 2025, though the records stayed publicly online for more than a year. The family says the ordeal cost $45,000 in lost wages and legal fees and traumatised the children. The sheriff's office did not respond to the Trib's questions.

High Unidentified facial recognition system

Phoenix: facial recognition match put Javier Lorenzano Nunez in jail for nearly a year for a 1998 murder; charges dismissed, federal suit filed

Javier Lorenzano Nunez was arrested in October 2024 for the 1998 murder of Sarah Carr after investigators ran the old suspect's MVD photo through facial recognition databases, got 250 possible matches and zeroed in on him — with Phoenix police publicizing the arrest in a press release and a special video. Charges were dismissed without prejudice in August 2025 after forensic evidence, including DNA and fingerprints, excluded him. His lawsuit alleges Phoenix police knew from a 2017 fingerprint analysis that his prints did not match, and ignored a 2007 Puerto Rico lead on the original suspect with the same name, date of birth and social security number. He spent nearly a year in jail. The federal suit against the Phoenix Police Department and the Maricopa County Attorney's Office alleges gross negligence, false arrest, false imprisonment and defamation.

High Unidentified facial recognition system

Maryland/Oklahoma: Kimberlee Williams, an Oklahoma woman who says she had never been to Maryland, was arrested in June 2021 and jailed about six months on bank-fraud charges in three Maryland counties after a facial-recognition search by an unknown CrimeDex respondent named her; all charges were dismissed by December 2021 and the ACLU filed complaints in April 2026

According to police and bank records reviewed by the Washington Post, in December 2019 a woman made fraudulent withdrawals at a SunTrust branch in Potomac, Maryland; branches in two other Maryland counties were also defrauded. A bank investigator posted the surveillance photos on CrimeDex, an online network of police and corporate investigators; an unknown respondent ran the image through facial-recognition software and returned Kimberlee Williams's name and photo, and the investigator told Montgomery County police that facial-recognition software had identified her. Police in Montgomery, Prince George's and Anne Arundel counties charged her in 2020 on the investigator's identification and their own visual comparison of photos (two departments also cited her old Oklahoma check-fraud convictions), without reported checks of her whereabouts; Montgomery County police did not disclose the facial-recognition search in the charging application, according to police records reviewed by the Washington Post and the ACLU. On 23 June 2021 she was arrested at the gate of Fort Sill near Lawton, Oklahoma, while accompanying her daughter on a DoorDash delivery, spent about three weeks in an Oklahoma jail, was flown to Maryland, and faced 16 charges including 12 felonies. The counties dismissed the charges one after another, in October and December 2021, after about six months in jail. Her daughter says she was depressed afterwards; the ACLU says the ordeal harmed her mental and physical health. On 14 April 2026 the ACLU and ACLU of Maryland sent complaint letters to the three departments seeking an apology and reforms.

Medium Unidentified facial recognition system

Southampton: Alvi Choudhury arrested at home on suspicion of a Milton Keynes burglary after a police retrospective facial recognition match, held nearly 10 hours, claiming damages

The Guardian, in a joint report with Liberty Investigates, reported on 25 February 2026 that Alvi Choudhury, a 26-year-old software engineer, was arrested at his home in Southampton in January 2026 on suspicion of a £3,000 burglary in Milton Keynes, about 100 miles away, and held in custody for nearly 10 hours. Thames Valley Police had used automated retrospective facial recognition software, which matched him with CCTV footage of the burglary suspect. Choudhury says the man in the footage looked about 10 years younger and had different features. Thames Valley Police wrote to him that the arrest may have been the result of bias within facial recognition technology, and told the Guardian that the decision rested on the investigating officers’ own visual assessment and was not influenced by racial profiling. Choudhury is claiming damages from Thames Valley Police and Hampshire Constabulary.

If you or someone you know is struggling, free and confidential support is available. Find a helpline near you at Signpost.