The perils of integrating AI in police operations
Kartavya Desk Staff
Source: IE
Subject: Governance
Context: In January 2026, the integration of Artificial Intelligence (AI) into Indian law enforcement has reached a critical milestone with the Delhi Police’s announcement of the Safe City Project and Maharashtra’s statewide rollout of MahaCrime OS AI.
About The perils of integrating AI in police operations:
Current AI Integration in India Policing:
• Delhi (Safe City Project): Launching in 2026, it features 10,000 AI cameras equipped with Face Recognition and Distress Detection (identifying screams or emergency gestures).
• Maharashtra (MahaCrime OS AI): An AI platform for predictive policing, aimed at identifying crime hotspots and processing complex investigative data.
• Surveillance Drones: Deployed for crowd and traffic management, providing a top-down view that replaces dozens of personnel on the ground.
• Data Backends: Systems like the CCTNS (Criminal Tracking Network and Systems) feed decades of historical data into these AI models to train them in pattern recognition.
Key Ethical and Administrative Concerns:
• Centralisation of Power: Policing is shifting from local beat cops to big data centres. This removes the human touch and makes it difficult for citizens to navigate a system where decisions are made by an invisible algorithm at the top.
• Excessive Policing & Imprisoning Cities: One AI camera is estimated to be as effective as 100 policemen. In cities like Hyderabad, with millions of cameras, the scale of surveillance creates a premise of suspicion where every citizen is a potential suspect.
• Historical Bias & Targeting: AI is trained on historical data. If past policing was biased against certain communities, the AI will learn to target those same groups, institutionalizing discrimination.
• Erosion of Fundamental Rights: AI tools can track protesters with ease, potentially chilling the Right to Dissent.
• Lack of Transparency: There is currently no AI Rulebook or statutory manual comparable to existing Police Manuals, leading to a black box where decisions cannot be easily challenged.
Challenges in the 2026 Landscape:
• Accuracy vs. Brutality: A tragic 2023 case in Telangana (the Khadeer Khan case) showed that reliance on grainy CCTV and facial recognition can lead to the detention and death of innocent people.
• Legal Vacuum: While the DPDPA (Digital Personal Data Protection Act) 2023 provides some safeguards, it contains broad exemptions for law enforcement, leaving a gap in protecting individual privacy against AI overreach.
• The Guilty until Proven Innocent Shift: Experts argue that AI-led policing flips the constitutional principle of the presumption of innocence by treating every public movement as data to be analyzed for anomalies.
The Way Ahead:
• Statutory Framework: Enacting specific laws for AI in policing that mandate Safety Tests and public disclosure of algorithmic logic before deployment.
• Human-in-the-Loop: AI must remain an assistive tool. Final decisions regarding arrests or detentions must always be made by a human officer held legally accountable for the action.
• Algorithmic Audits: Regular, third-party audits of police AI to detect and remove caste, religious, or gender-based biases.
• Police Reforms: Reforming acts like the Criminal Procedure (Identification) Act, 2022 to ensure that data collection of non-convicts is strictly limited and proportional.
Conclusion:
Technology is not a substitute for institutional integrity. To prevent AI from becoming a tool of digital authoritarianism, India must ensure that its march toward a high-tech future remains anchored in Constitutional Values. A safer world is created not by watching every citizen, but by building a society rooted in trust, transparency, and the Rule of Law.
Q. “The integration of artificial intelligence into policing marks a shift from community-based law enforcement to centralised algorithmic control”. Evaluate the nature of this shift. Analyse its implications for police accountability and its impact on democratic freedoms. (15 M)