GS Papers: GS2, GS3
What Happened
India's major cities are increasingly deploying AI-powered policing and traffic management systems, transforming law enforcement from reactive to data-driven and predictive. Delhi, Bengaluru, Hyderabad, and Mumbai lead these pilots, with expansion planned across more cities.
Bengaluru's ASTraM (Actionable Intelligence for Sustainable Traffic Management) system ingests live feeds from CCTV cameras, Automatic Number Plate Recognition (ANPR) systems, and open data sources, with AI engines identifying patterns in recurring and non-recurring congestion. Delhi's Safe City Project, launching in 2026, features 10,000 AI-enabled cameras with face recognition and distress detection capabilities. Maharashtra is implementing MahaCrime OS AI — a predictive policing platform for identifying crime hotspots and processing investigative data.
Cities including Kalyan-Dombivli, Pimpri-Chinchwad, New Town Kolkata, Varanasi, and Visakhapatnam have embedded AI into civic infrastructure, deploying computer vision, predictive mapping, and intelligent alerts. While proponents argue these systems reduce congestion and improve public safety response, critics — including human rights organisations and digital rights groups — raise serious concerns about algorithmic bias, erosion of privacy, and the creation of surveillance infrastructure without adequate legal frameworks.
Static Topic Bridges
1. AI-Powered Traffic Management — Technology and Scale
The core of AI traffic management systems in India involves multi-source data integration: CCTV feeds, ANPR cameras (which read vehicle licence plates), GPS data from connected vehicles, and operator-reported inputs. Machine learning models process this data in near-real-time to predict congestion, detect incidents (accidents, wrong-way driving, stalled vehicles), and adjust signal timings dynamically.
Bengaluru's ASTraM exemplifies this approach. Traditional traffic management was reactive — a signal jam was addressed after it formed. ASTraM shifts to predictive intervention: identifying formation patterns and re-routing before congestion fully builds. Integration with ANPR enables automated challan generation for traffic violations, reducing the manual enforcement burden. These systems are positioned under the Smart Cities Mission — an urban infrastructure programme launched in 2015, with 100 cities selected for funding — as part of the Integrated Command and Control Centres (ICCC) that cities are required to establish.
2. Predictive Policing — How It Works and What It Predicts
Predictive policing uses historical crime data, geographic information, demographic data, and socio-economic indicators to identify "crime hotspots" — locations or populations that algorithms flag as higher-risk for future criminal activity. Two broad categories exist: place-based prediction (where crime might occur) and person-based prediction (who might commit or be a victim of crime).
Maharashtra's MahaCrime OS integrates first information report (FIR) databases, criminal records, and geographic crime data to generate risk scores by location and suggest preventive patrol deployment. Delhi's project adds facial recognition in public spaces — a significantly more intrusive tool, capable of identifying individuals from camera feeds without their knowledge.
Advocates argue this improves resource allocation and allows preventive intervention. Critics note that predictions are only as good as historical data — and historical data reflects existing policing biases, generating self-reinforcing feedback loops where already over-policed communities continue to be flagged as high-risk regardless of actual crime rates.
3. Constitutional and Rights Concerns — Privacy, Article 21, and DPDP Act
The Supreme Court's nine-judge bench in Justice K.S. Puttaswamy v. Union of India (2017) recognised the right to privacy as a fundamental right under Article 21. Mass surveillance systems — CCTV networks, facial recognition, ANPR — constitute an ongoing intrusion into this right. For such intrusions to be constitutionally valid, they must satisfy the triple test: legality (authorised by law), necessity (serves a legitimate state aim), and proportionality (minimally restrictive means used).
India's Digital Personal Data Protection (DPDP) Act, 2023 establishes data protection rights but contains broad exemptions for state security and law enforcement. Currently, no specific legislation governs the development, procurement, or deployment of AI policing tools. The legal vacuum creates a situation where cities can deploy facial recognition and predictive analytics without statutory authorisation, judicial oversight, or mandatory impact assessment.
The LSE Human Rights Programme and other researchers have highlighted that predictive policing tools in India disproportionately target Muslims, Dalits, and other marginalised communities — because historical policing data itself reflects caste and communal bias. Without algorithmic auditing requirements or transparency mandates, citizens have no mechanism to challenge predictions made about them.
4. Data Governance and the Smart Cities Framework
The Smart Cities Mission's Integrated Command and Control Centres aggregate data from across city systems — transport, utilities, public safety, emergency services — into centralised dashboards. This creates both operational efficiency and novel data governance challenges: who owns the data collected, how long it is retained, who can access it, and under what legal authority can it be shared with police or other agencies?
India's current framework does not have answers codified in law. The DPDP Act focuses on processing of personal data by "Data Fiduciaries" — primarily private entities — and its application to state agencies remains ambiguous. The absence of a dedicated AI regulatory framework (India's approach is currently advisory, through the IndiaAI governance guidelines) means accountability for AI policing tools rests primarily with police commissioners and municipal corporations — bodies without expertise in algorithmic accountability.
Key Facts and Data
- Bengaluru ASTraM: AI traffic system integrating CCTV, ANPR, and open data; identifies congestion patterns predictively
- Delhi Safe City Project: 10,000 AI cameras with face recognition and distress detection (2026 launch)
- Maharashtra MahaCrime OS: AI platform for predictive policing and crime hotspot analysis
- Smart Cities Mission: 100 cities selected; all required to establish ICCCs
- Legal basis for AI policing: no specific statute — existing police acts and general administrative law
- Puttaswamy judgment (2017): right to privacy is a fundamental right under Article 21
- DPDP Act, 2023: data protection law but with broad state security exemptions
- India AI Governance Guidelines: advisory, not legally binding; no AI-specific law enacted
- Key concern: algorithmic bias creating discriminatory feedback loops in policing
- No mandatory algorithmic impact assessment or auditing requirement for public AI systems in India