A retired senior official from Israel’s Shin Bet and Israel Police has proposed what could be the next frontier in AI-powered public safety: Real-Time Prevention Centers that connect scattered data across police, hospitals, municipalities, and social services to identify crime risks before they produce victims.
🔍 THE BOTTOM LINE
The idea is seductive — connect fragmented data streams, let AI find patterns no single agency can see, and prevent crime before it happens. But the same technology that connects hospital admissions to streetlight outages to police reports also creates a surveillance infrastructure with obvious civil rights implications. The safeguards matter more than the algorithms.
The Proposal
Writing in YNet News on August 3, Major General (ret.) Boaz Gilad argues that current Real-Time Crime Centers — already operational in many cities — remain fundamentally reactive. An incident occurs, information is verified, resources are deployed. By then, the public has already been exposed to harm.
His proposed Real-Time Prevention Centers (RTPCs) would connect data that currently sits in institutional silos:
- Police: intelligence and crime data
- Hospitals: violence-related injury admissions
- Municipalities: infrastructure failures, lighting issues
- Schools and social services: indicators of growing distress
- Transportation systems: unusual movement patterns
- Residents: community feedback and reporting
The AI layer would connect seemingly unrelated signals. A series of minor disturbances, recurring business complaints, deteriorating street lighting, and a rise in emergency-room admissions may look insignificant separately. Together, they could constitute a warning that violence is escalating in a specific area.
The Entertainment District Example
Gilad offers a concrete scenario. Consider an entertainment district where assaults are rising. The traditional response is to add patrols after repeated incidents. A prevention-oriented approach would connect police reports with hospital admissions, traffic patterns, business complaints, and environmental conditions.
It might reveal that violence peaks when venues close, public transportation stops, and lighting is poor. The answer could combine targeted police presence, extended transportation hours, better lighting, alcohol enforcement, temporary cameras, and coordination with venue owners. No single measure is revolutionary — their synchronised use before the next incident is.
The Civil Rights Question
The proposal is careful to include safeguards: defined legal authority, strict access controls, auditable algorithms, human validation, transparent oversight, and active testing for bias and false positives. Gilad explicitly states that “prevention cannot be built at the expense of civil rights, because without public trust the information and cooperation on which the model depends will disappear.”
But civil liberties organisations have been warning about this trajectory for years. The Brennan Center for Justice has documented how unregulated AI in policing can amplify existing biases, particularly against minority communities. The concern is not theoretical — predictive policing systems in the US have already demonstrated disproportionate impacts on communities of colour.
The core tension: the same data integration that identifies a dangerous intersection can also identify a dangerous person. The line between “identifying conditions” and “predicting individuals” is thinner than any algorithm can reliably maintain.
The Global Pattern
Gilad frames this as a global shift, not an Israeli initiative. The obstacle, he writes, “is not technology; it is mindset.” Public safety organisations remain structured to respond within institutional boundaries. Real-time prevention requires them to share information, define joint responsibility, and act as one network around a common risk.
The infrastructure is already being built. Deloitte’s analysis of surveillance and predictive policing through AI notes that the technical capabilities exist in most modern cities. What is missing is the legal framework and institutional willingness to share data across agency boundaries.
New Zealand is not immune to this trend. The Privacy Commissioner has already flagged concerns about AI surveillance expansion, and the government’s encryption backdoor proposal demonstrated that data-sharing ambitions can outpace privacy safeguards.
What Makes This Different From Predictive Policing
Gilad is careful to distinguish his proposal from the predictive policing models that have drawn civil rights challenges. The goal, he writes, “is not to predict people, but to identify dangerous conditions before they produce victims.” The system should trigger human assessment and a proportionate preventive response — not automatic police action.
This is a meaningful distinction. Predictive policing systems like PredPol and COMPASS attempt to forecast where crimes will occur or who will commit them. RTPCs, at least in theory, identify environmental conditions that correlate with crime — broken lighting, transport gaps, hospital admission spikes — and route resources to fix the conditions, not arrest the people.
Whether that distinction holds in practice depends entirely on the safeguards. A system designed to fix streetlights can be repurposed to track individuals in the same building. The audit trail, access controls, and independent oversight are what prevent mission creep — not the algorithm itself.
❓ FAQ
Is this the same as “Minority Report” predictive policing? No — at least in theory. The proposal targets environmental conditions (lighting, transport gaps, hospital admissions) rather than individuals. The distinction matters but depends on safeguards to prevent mission creep.
Does New Zealand have anything like this? Not yet at this scale. NZ police use some data-driven approaches, but the integrated cross-agency model Gilad describes does not exist in Aotearoa. The Privacy Act 2020 and the government’s AI framework would both apply if such a system were proposed.
What are the main civil rights concerns? Data sharing across agencies creates a surveillance infrastructure that can be repurposed. Bias in algorithms can disproportionately flag minority communities. And the line between “identifying conditions” and “predicting individuals” is difficult to maintain in practice.
Who oversees AI in policing globally? It varies. The EU AI Act classifies law enforcement AI as high-risk with mandatory audits. The US has no federal framework — 14 states have their own rules. The UK’s pending AI Safety Bill grants inspection powers. China regulates through the CAC.
🔍 THE BOTTOM LINE
The data exists. The AI exists. The institutional silos are the obstacle. A retired police chief has sketched a vision of integrated public safety that could save lives — or normalise a level of cross-agency surveillance that no democratic society has ever sanctioned. The technology is not the hard part. The safeguards are.
📰 Sources
- YNet News — The next policing revolution is preventing crime before it happens
- Brennan Center — Dangers of Unregulated AI in Policing
- National Law Review — AI Police Surveillance Bias
- Deloitte — Surveillance and Predictive Policing Through AI
- Singularity.Kiwi — NZ Government Surveillance Expansion
— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.
