Speed and accuracy are different things. We've built systems that are very fast at being wrong.
Across the world, police forces are turning to artificial intelligence not as a novelty but as a lifeline — overwhelmed by oceans of data, shrinking budgets, and crimes that move faster than human investigators can follow. From London's Metropolitan Police using Palantir's Nectar system to expose misconduct within its own ranks, to Indian cities preparing AI surveillance grids for tens of millions of festival-goers, the machinery of law enforcement is being quietly rewired. The promise is precision and speed; the peril is that systems trained on flawed human judgments may simply accelerate the oldest injustices.
- Police departments from London to Kerala are deploying AI tools that can do in seconds what once took detectives months — scanning jail calls, dark web data, and thousands of CCTV feeds simultaneously.
- The UK's Metropolitan Police used Palantir's Nectar AI to investigate its own officers, producing three arrests and 98 under review within a single week — a speed that unsettled as much as it impressed.
- India is weaving 17,000 police stations, courts, prisons, and forensic labs into a single AI-readable data pool, while cities like Nashik are building surveillance systems to manage 120 million visitors with feeds from 5,000 cameras and drone networks.
- Australia's experience offers a stark warning: 60% of AI-issued traffic fines were overturned after the system mistook wallets and glasses for mobile phones, exposing how automation bias can industrialize error.
- The deepest tension is not technical but moral — whether societies are building tools that deliver justice more swiftly, or tools that simply make discrimination and wrongful arrest faster and harder to challenge.
In April, London's Metropolitan Police deployed an AI system called Nectar, built by Palantir, to investigate its own officers. Within a week, it had flagged hundreds for violations ranging from roster manipulation to sexual assault. Three were arrested; ninety-eight placed under review. The tool works by running algorithms across internal databases, hunting for patterns that don't belong — and it found them with unsettling speed.
This is part of a broader British initiative: a £115 million AI Centre rolling out tools across 43 police forces. The logic is one of necessity, not curiosity. Police departments worldwide are drowning in data — footage from body cameras, drones, social media, forensic logs — and no human team can process it fast enough. Machines are being asked to swim where humans sink.
The applications are wide-ranging. During the Maha Kumbh festival in Uttar Pradesh, AI software monitored thousands of live CCTV feeds for overcrowding and altercations in real time. Nashik, expecting 120 million visitors in 2027, is building a system drawing on feeds from over 5,000 cameras, drones, and IoT sensors. India's home ministry is connecting 17,000 police stations into a single integrated platform linking police, courts, prisons, and forensics.
In Anchorage, Alaska, AI helped detectives crack cold cases by scanning jail calls, social media, and old case files across languages in seconds. In Kerala, police used an AI tool to trace child sexual abuse material across Facebook, Telegram, and Instagram, leading to an arrest within weeks. INTERPOL used AI analytics in Operation Shadow Storm to track global scam networks, predicting where stolen funds would flow and freezing accounts across borders.
Yet the dangers are as real as the promise. Australia's AI traffic enforcement system issued 300 fines per day — and 60% were later overturned after the software mistook wallets and glasses for phones. Maharashtra is building a tool to identify suspected illegal immigrants by analyzing speech patterns, raising the question of whether it can distinguish dialects or will simply arrest people who sound unfamiliar. Faulty training data, automation bias, and human prejudice encoded into algorithms risk not just inefficiency, but wrongful arrests and the systematic targeting of marginalized communities. The question haunting every deployment is whether these systems deliver justice faster — or simply make old errors harder to see and harder to fight.
In April, the Metropolitan Police in London deployed an artificial intelligence system called Nectar, built by the American intelligence firm Palantir, to investigate its own officers. Within a single week, the software had flagged hundreds of police personnel for violations ranging from roster manipulation and workplace misconduct to corruption and sexual assault. Three officers were arrested. Ninety-eight more were placed under assessment. The tool works by running algorithms across internal police databases, searching for patterns and anomalies—the things that don't belong. It found them with unsettling speed and precision.
This deployment is part of a larger British initiative announced earlier this year: a £115 million AI Centre designed to roll out artificial intelligence across 43 police forces in England and Wales. The rationale is straightforward. Modern policing faces mounting pressure. Digital crime crosses borders. Financial fraud grows more sophisticated. Budgets shrink. Police departments worldwide—from India to South America to the United States—are turning to AI not out of curiosity but out of necessity. They are drowning in data and need machines to help them swim.
The first wave of police AI adoption came through facial recognition technology. These systems run on top of existing CCTV feeds, identifying faces and objects in crowds. A detective can ask the software to locate a man in an orange kurta or a blue car across weeks of footage, work that once consumed days of human attention. During the Maha Kumbh festival in Uttar Pradesh in 2025, police deployed video analysis software from the Gurgaon startup Staqu Technologies to monitor live CCTV feeds for overcrowding, fire, and altercations, allowing them to respond in real time. Nashik, preparing to host the Simhastha Kumbh Mela in 2027 and expecting 120 million visitors, convened an AI strategy workshop with MIT, Meta, Google, and Microsoft to build a system capable of processing feeds from more than 5,000 CCTV cameras, IoT sensors, drones, and mobile networks simultaneously.
The core problem these systems address is not a shortage of evidence but an abundance of it. Police departments now collect footage from security cameras, drones, body cameras, forensic logs, and social media. No human team can sift through 5,000 camera feeds during a festival. Machines can. Police are increasingly handing forensic analysis to AI, which builds patterns and identifies crime hotspots and high-risk individuals—a practice called predictive policing. In India, the Union home ministry is implementing CCTNS 2.0, a centralized platform connecting 17,000 police stations across the country, part of a larger system called the Inter-operable Criminal Justice System that links police, courts, prisons, and forensics into a single data pool. Run AI tools on this integrated database, the ministry hopes, and policing becomes faster and more precise.
The results in some cases have been striking. In Anchorage, Alaska, the police department adopted technology from a startup called Closure that searches large datasets for evidence in unsolved cases. The software analyzed jail calls, interviews, social media, photographs, and old case files, flagging new leads and crucial messages even across different languages. Police Chief Sean Case noted that detectives sometimes spend over 1,000 hours listening to jail calls searching for a single word or phrase. The AI does this work in seconds. Kerala Police used an AI tool called Katalyst to investigate child sexual abuse material found on the dark web. The software helped officers narrow down a suspect by analyzing photos and data scattered across Facebook, Telegram, and Instagram. Within weeks, they arrested a woman in Thiruvananthapuram who was posting images of her niece. "This problem was amplified by technology," said Ankit Asokan, the state's cyber crime superintendent. "We must use technology to fight back."
AI has also proven useful in tracking financial crime. INTERPOL launched Operation Shadow Storm in March, using AI-driven data analytics to follow the money trails of global scam networks. The system can predict where stolen funds will flow within seconds of a transaction and freeze accounts across borders. India's home ministry is working with IIT Bombay and the Reserve Bank of India to develop a model that identifies suspect mule accounts in real time, flagging fraudulent transactions before they complete.
Yet the same power that makes these systems useful also makes them dangerous. The Maharashtra government is building a ₹3 crore AI tool intended to identify suspected illegal Bangladeshi nationals by analyzing speech patterns, tone, and linguistic usage. The question is whether the software can distinguish between a West Bengal dialect and a Bangladeshi one—or whether it will simply arrest people who sound wrong. In Australia, police used CCTV to automatically detect seatbelt violations and phone use while driving, issuing 300 fines per day based on the AI's determinations. Sixty percent of those fines were later overturned, many because the system had flagged wallets, glasses, or battery packs as phones. Automation bias, faulty datasets, and human prejudice baked into training data can produce cascades of false positives. The risk is not merely inefficiency. It is the arrest of innocent people, the targeting of marginalized communities, and the normalization of surveillance based on algorithmic error. As the technology spreads, so does the question of whether we have built a system that works, or one that simply works faster at being wrong.
Citas Notables
This problem was amplified by technology. We must use technology to fight back.— Ankit Asokan, SP Cyber Crime, Kerala Police
When we tested the software, one of the things we primarily used it for was throwing in jail call data.— Police Chief Sean Case, Anchorage Police Department
La Conversación del Hearth Otra perspectiva de la historia
Why does the UK need an AI system to investigate its own police officers? Isn't that what internal affairs already does?
Internal affairs works, but it's slow and it's human. A team of investigators can only look at so many cases at once. Nectar runs through years of internal data in days, spotting patterns a person might miss—officers who consistently abuse shift rosters, or whose names appear in misconduct reports across different precincts. It's not replacing judgment; it's accelerating the search.
And it found hundreds of violations in a week. That's either brilliant or terrifying.
Both. The speed is real. But it also raises a question: were those violations always there, just invisible? Or is the AI flagging things that wouldn't have been flagged before? Three arrests and 98 assessments is significant, but we don't yet know how many of those assessments will lead anywhere.
Let's talk about the Australian traffic fines. Sixty percent overturned. How does an AI system get something that wrong?
The system was trained to recognize phones in hands. But a phone in a hand looks similar to a wallet, a glasses case, a battery pack. The AI learned a pattern, but the pattern was incomplete. It saw the shape and made a guess. When you're issuing 300 fines a day, those guesses add up fast.
So the problem is the training data, or the design of the system?
Both. But also the assumption that because a machine is fast, it must be accurate. Speed and accuracy are different things. We've built systems that are very fast at being wrong.
The Maharashtra tool that analyzes speech to identify illegal immigrants—that feels like it's asking for trouble.
It is. Language and dialect don't map cleanly onto nationality. A tool trained on limited data will make crude distinctions. And once it starts flagging people, those flags become part of their record, whether they're accurate or not. The speed of the system becomes a liability.