"Intelligence infrastructure" is used loosely, and it risks becoming one more label for anything that combines data and AI. It means something specific, and the difference matters to anyone deciding what to buy.
Intelligence infrastructure is a persistent, structured and evidenced foundation of intelligence that people, applications and AI agents can query directly. The evidence is assembled before the question is asked. Reports are one output of it, not the product itself. Or, more simply: intelligence you can query, not just reports you receive.
Several useful things are often mistaken for it
AI for insurance. The phrase covers everything from claims chatbots to submission scrapers. Most of these automate a single task. Adding a model to one step of a process does not change where the evidence comes from or whether anyone can check it.
More or better reports. A detailed country assessment is valuable. Delivered as a long document in the middle of an event, it also adds to the reading. Better reports improve quality without reducing the work of turning them into a decision.
Process automation. Workflow tools move data between systems and route approvals. They are good at it. They cannot make uncertain or unstructured information trustworthy, and automating a process that runs on unchecked inputs only moves those inputs faster.
Dashboards of history. Loss histories and portfolio dashboards explain what happened in earlier years and are essential for planning. They are not built to answer what is happening around a particular site this week, and why.
Consultancy through a portal. Putting expert research online improves access to it. Where each answer is still researched and written by hand on request, capacity still depends on people.
It has four defining characteristics
A foundation built before the question. CORTEX combines current and historical intelligence on geopolitical and security developments, disruptive incidents and natural hazards with hundreds of contextual and reference datasets, including critical infrastructure, trade and transport, environmental conditions, media coverage and government advisories. AI and data science structure and enrich it continuously, so the groundwork is done before anyone needs it.
Direct access for people, systems and AI agents. Analysts can query it, applications can draw on it through standard interfaces, and AI agents can use it as their source of facts. They all draw on the same evidence, so different questions and different tools start from the same place.
Repeatable production and comparison. The same picture can be produced again, on a schedule, and compared with the last one. That shows what has materially changed without rebuilding the research each time.
Evidence that can be examined. Sources, dates, methods and known limitations stay available beneath every output.
Three commitments make it trustworthy
Infrastructure that informs consequential decisions has to earn trust, and we hold CORTEX to three commitments.
The first is to show the evidence. Every finding should lead back to its sources, its dates and the method behind it. Source facts, computed results, model-assisted classifications and human conclusions are different kinds of statement, and they should be labelled as such. A correlation is a lead to investigate, not proof of a cause.
The second is to say where coverage runs out. No foundation sees everywhere equally. Gaps in reporting, delays and unconfirmed material should be stated, so that decision-makers know where the evidence ends and uncertainty begins.
The third is to leave the judgement with people. Technology carries the research burden. Analysts, advisers and decision-makers remain responsible for interpretation and for the decision itself.
It changes what AI can safely do
A general-purpose language model can write a fluent account of a risk situation without establishing which sources, figures or dates support it. Asked again, it may produce a different account. That is a poor basis for a decision someone may later have to explain.
Intelligence infrastructure separates the evidence from the question. The facts, computations and history sit in the foundation; the model or the analyst works from them and can point back to them. The question can change. The evidence should not have to be invented again. Generated interpretation still needs checking, but it is checked against something that exists.
Five questions tell the difference
When assessing anything sold under this name, five questions separate infrastructure from a useful tool:
- Is the evidence assembled before you ask, or researched after you ask?
- Can you see the sources, dates and methods behind every output?
- Does it tell you where its coverage is weak?
- Can your people, your systems and your AI tools all query the same foundation?
- Can it show what has changed since last time without the research being done again?
A product that answers yes to all five is infrastructure. One that answers yes to some may still be worth having, but the research burden will still sit with your people.
The next article in this series asks whether insurers should build this capability themselves.
First published in an earlier form in early 2026.