The use of facial recognition technology in policing public protests raises questions of accuracy, privacy and proportionality. Discuss with reference to recent developments.

Facial Recognition Technology (FRT) matches live camera images against stored biometric databases. In July 2026, Delhi Police deployed FRT vans at the Jantar Mantar protest site and reportedly identified over 2,500 people with criminal records [1]. Police surveillance of an assembly protected by Article 19(1)(b) therefore needs a firm legal basis and a reliable tool.

Accuracy: a tool prone to error

• False positives: the system placed at least 25 people at the protest who prison and court records show were in jail at the time [2].
• Algorithmic bias: NITI Aayog has flagged inaccuracy, misidentification and dataset bias as core risks of FRT [3].
• Consequence: innocent people may be wrongly linked to unrest and brought into criminal proceedings.

Privacy: biometric data with weak guardrails

• Fundamental right: Puttaswamy (2017) recognised privacy as part of Article 21 [4]. A face is sensitive, permanent biometric data.
• Statutory gaps: the DPDP Act, 2023 allows the government to exempt state agencies for security and public order [5]. No dedicated law governs how police use FRT.
• Chilling effect: if attending a protest means being logged by the state, people may stop exercising their right to dissent.

Proportionality: the constitutional test

• Puttaswamy requires legality, a legitimate aim, necessity and procedural safeguards before privacy can be restricted [4].
• Scanning an entire crowd to find a few suspects is hard to justify as the least restrictive means, particularly with a tool that has shown errors [2].

The case for FRT

• It can help locate wanted offenders and missing persons quickly, which is a legitimate aim if it is used narrowly and checked by a person [3].

Way forward

• A specific enabling law with independent oversight, as NITI Aayog recommends [3].
• Mandatory accuracy audits, human verification before any police action, and limits on how long data is kept.
• Clear purpose limits, so that FRT is used only to find named suspects and not to watch everyone in a crowd.

FRT can help policing, but recent cases show it is error-prone and lacks a clear legal framework. If it rests on a dedicated statute, is audited, and is held to the Puttaswamy proportionality test, India can achieve Responsible AI for All while protecting free speech, the right to assemble and dignity.