AI Tradecraft for Executive Protection

Executive Summary

Close protection has always been a discipline of advance work. The detail leader who knows the route, the venue, the local rhythm, and the threat picture before the principal arrives is the one who keeps the principal safe. The tradecraft hasn’t changed in fifty years. The tools have.

The change underway is not “more data.” Every Global Security Operations Center (GSOC) already has more data than it can read. The hard part — and the part that has only recently become solvable — is turning the world’s flood of raw reporting into a usable picture: reading thousands of articles a day in more than 150 languages, watching hundreds of channels across the messaging apps that matter locally, and connecting a road closure to a planned protest to a threat posted six months ago. Then producing a brief a detail leader can act on, in plain English, in under a minute.

That synthesis layer is what GeoSure built. We call it GRO — the GeoSure Global Risk Observatory. It runs continuously: it reads the world’s public reporting, pins each item to the neighborhood it happened in, combines it with trusted ground-truth data, and hands the result to a series of AI assistants that can answer questions about it on demand.

The discipline that comes out of this — call it AI tradecraft — is not a replacement for trained people in the field. It is a permanent, tireless intelligence layer behind every advance and every movement. It shifts the unit of attention from “the city we’re going to” to “the small area around the principal in the next 60 minutes.” And it can be combined with the data your security program already owns — access control, corporate travel, vehicle tracking, existing intel subscriptions — into one coherent picture. The technology to do this exists today. What’s required is implementation.

This paper describes how that layer works, what it can and cannot do today, and how it fits the protective cycle for corporate executives and the principals a security program protects. It is grounded in what GRO actually does in production — with honest limits.

Key Points

An always-on intelligence layer with neighborhood-level resolution can strengthen every stage of the protective cycle.
A new generation of AI assistants can reason about a location, not just return a list of articles — the difference between “here are 40 articles” and “here is what changed along the principal’s route in the last hour, and what it means.”
Open-source intelligence is fragmented by geography. The platform that matters in Riyadh is not the one that matters in Bogotá or Jakarta. A credible program is designed around that fact.
The synthesis layer is the advantage, not the raw data. The major datasets are available to anyone with a budget. What separates a usable picture from noise is the layer that filters, locates, de-duplicates, scores, and summarizes.
Your internal data belongs in the picture. GRO is built to fold in your own feeds — badge access, travel, vehicle GPS, internal intel — under your control.
The right posture is public sources only for the open layer, customer-controlled access for your private data, persistent monitoring, and full logging. This is not a surveillance product. It is an advance team that never sleeps.

1. The Intelligence Gap in Modern Close Protection

The classic EP intelligence model rests on three tiers, each with a known limitation:

Between operational and tactical sits a 30-to-180-minute window where modern threats now organize — protests forming on messaging apps, paparazzi tipped off on social video, opportunistic crowds spawning in group chats, road closures cascading through local social media. Hostile surveillance patterns often show up in foreign-language posts before they ever appear in a route survey. Civil disturbance, infrastructure failure, transit strikes, and weather can all shift the picture during a movement.

A traditional EP intelligence analyst can fill some of this gap — but typically for one principal, in one language, for a few hours a day. That doesn’t scale for a Fortune 500 with executives moving across 30 countries, or for a security program covering multiple principals, offices, and dozens of annual destinations.

This is the gap AI tradecraft closes.

2. What “AI Tradecraft” Actually Means

We define AI tradecraft as the disciplined use of AI assistants that can independently look things up, reason about what they find, and produce a usable written output — applied to the protective intelligence cycle.

Three things separate this from “we use ChatGPT in the GSOC”:

1. Purpose-built data, not the open web

A general chatbot searches Google. An AI tradecraft system queries a curated, scored, geographically organized library of safety, conflict, news, and open-source data — refreshed continuously and pinned to the neighborhood, not just the city.

2. An assistant that does the work, not a dashboard you read

Instead of you pulling up a dashboard and interpreting it, a capable AI assistant can do the legwork itself: look up a location, pull its safety scores, check live news around it, compare it to historical patterns, and draft a brief. The detail leader doesn’t query a database — the detail leader asks a question, and the assistant does the work.

3. Persistent monitoring, not occasional check-ins

A good analyst checks the picture every few hours. An AI assistant checks it every few minutes — on every facility, every residence, and every active principal location at once, in every relevant local language. When the picture changes in a way that matters, the GSOC is told. When nothing changes, no one is interrupted.

This is not a future state. It is in production at GeoSure today — serving an Asia-Pacific travel platform with millions of monthly users, and a corporate pilot pipeline monitoring hundreds of office and meeting locations worldwide in real time.

3. The System — GRO and the GeoSure Platform

GeoSure spent twelve years building the underlying data foundation and the past year building the AI layer that reasons over it. The system that holds it together is GRO — the GeoSure Global Risk Observatory. It has four parts that matter to a protection program: a geographic foundation, a continuous reading-and-synthesis pipeline, an open-source layer that respects how the world actually posts, and a way to fold in your own internal data.

3.1 Geographic foundation — the neighborhood as the unit of work

The world is not made of cities. It is made of neighborhoods, corridors, blocks, and addresses. GRO divides the populated surface of the planet into roughly five million small cells, each about three-quarters of a square kilometer — about the size of a city neighborhood. Covering over 850,000 populated neighborhoods in 200 countries.

Every cell carries a safety score on a 1-to-100 scale across six separate dimensions — physical violence, theft, health and medical, political freedoms, women’s safety, and a day-versus-night difference — rather than collapsing everything into a single number. Behind each cell sit authoritative inputs: verified ground-truth event data, global governance and prosperity indices, national and local crime statistics where they exist, and a database of over 71 million named places.

For the protection team, this means:

A route can be scored block-by-block, not city-by-city.
A residence assessment isn’t a country-level briefing — it’s the neighborhood around the home.
An office and the route between office and residence become first-priority focus, not afterthoughts.

3.2 Reading the world, continuously

GRO continuously reads the world’s publicly reported events. The volumes are intentionally large, but the volume is not the point. The point is what happens between collection and the desk of the protection director.

On the input side, GRO pulls from over 10,000 curated news feeds, commercial newswires, and a real-time global event stream — on the order of 100,000 articles a day, across more than 150 languages, in effectively every country.

Raw collection is the cheap part. The valuable part is what GRO does next:

Safety screening. Every article is read and judged for whether it represents a real, immediate safety signal for people on the ground. Most are filtered out — that’s the point. Volume becomes signal.
Location pinning. Each surviving article is tied to a precise point and to the neighborhood it happened in. An incident “near the central market in Lagos” becomes an event tied to the right neighborhood, not just the city.
De-duplication. Forty wire-service versions of the same event are recognized as one event — including when it’s reported in three different languages.
Synthesis. Related events are turned into analyst-grade summaries: what happened, where, when, and in what context.

The result is a few thousand fully-located, de-duplicated, classified items a day — each tied to a neighborhood and ready for a person or an assistant to use. For live sources, the time from original publication to a located event in GRO is measured in minutes, not hours. This live layer is paired with trusted historical ground-truth data to provide both the current picture and the longer-term pattern.

3.3 Open-source intelligence is fragmented by geography

This is the part that surprises new corporate clients. Open-source intelligence is not one discipline — it is radically different by region. Where people live determines which platforms they post to about a road closure, a protest, a celebrity sighting, or a fight outside a bar.

A program built only around X and Reddit is functionally blind in two-thirds of the world. One that adds TikTok but ignores Telegram will miss the meaningful workforce signals in oil-and-gas operations across the Caspian, North Africa, and the Middle East.

3.4 Folding in your own data — access control, travel, telematics

This is the part most prospective customers don’t expect. A corporate GSOC already runs on a stack of internal systems — none of which talk to each other:

Access control and badge data (e.g., Lenel, CCURE, Brivo, Genea).
Corporate travel management (e.g., Concur, Amex GBT) — which knows where every executive is supposed to be, and when.
Vehicle tracking and motorcade GPS.
Existing commercial threat-intel feeds (e.g., Dataminr, Ontic, Crisis24).
Internal incident management and executive calendars, offices, and vehicle inventories.

The access-control system sees badge swipes. The travel manager sees itineraries. The analyst sees public news. None of them sees the combined picture — and the combined picture is what determines whether a threat is caught before it becomes an incident.

GRO is designed to be the layer that pulls these together. The same system that reads public news and open-source data can read your private feeds — under your access controls, on your compliance terms, inside your own environment — and reason across both the public and private picture together.

Importantly, this is wired in carefully and conservatively. For access control, GRO consumes badge events — not personal identity details beyond what you choose to forward. For travel, itineraries automatically switch on monitoring for the relevant locations. For vehicle tracking, live position updates let alerts re-score against the route being driven.

The technology to do all of this exists today. There is no missing science. What’s required is implementation — building the connectors and deploying the system in your environment under your governance. That is an engineering engagement of weeks to months, not a flip-the-switch purchase, and we are clear about that scope from the first conversation. The line stays exactly where it should: public sources follow public-source rules; your private data stays under your control, on your compliance regime, inside your boundary, with full audit logging.

4. Limitations

No platform has every open source. Instagram, closed WhatsApp groups, Discord, dark-web forums, and (for now) TikTok are not yet in active production. The platforms that matter depend on geography, language, and the threat. A serious program names its blind spots and makes risk-informed choices.
Some trusted datasets lag reality. The most rigorous historical datasets rely on human verification and are therefore not real-time. GRO supplements them with live news and open-source layers, but a lag in the most carefully verified sources is a feature of the discipline, not something we can engineer away.
AI assistants make mistakes. They occasionally miscategorize or get something wrong. The right posture is AI-drafted, human-reviewed for any consequential decision (re-route, abort, evacuate). The assistant is a tireless senior intern, not an autonomous detail leader.
Speed is “minutes,” not “seconds.” A truly time-critical event — active shooter, vehicle ramming — will reach the GSOC through 911 and on-site security before any open-source signal does. AI tradecraft is for the 30-to-180-minute window. Tactical real-time remains the domain of the agent in the car.
Coverage is uneven by country. Ground-truth event data is strong globally; crime data is strong in developed countries and patchy elsewhere; news coverage is best in English and a handful of major languages and degrades from there.
Integration takes real work. Wiring in access control, travel, and telematics is straightforward but not free — it’s an engineering engagement, and we say so up front.
The protective cycle still requires humans. Always. The point is not to remove the detail manager. The point is to give security professionals the intelligence picture an entire analytical staff once provided — in their pocket, in real time.

AI Tradecraft for Executive Protection

How Always-On, Neighborhood-Level Intelligence Is Reshaping Close Protection for Corporate Executives