AI in Intelligence Analysis: The Reading The Room Report

11^th Aug 2026 By Duane Rivett – VP & Co-Founder

The Reading the Room report examines a threat that isn’t obvious at first. It isn’t that AI produces bad intelligence. It’s that AI produces intelligence fast enough, and convincingly enough, that no one stops to check. The bigger shift isn’t happening at the point of production. It’s happening at the point of consumption, in how the people receiving intelligence now expect to read it, question it and act on it.

The analyst isn’t obsolete. The old workflow is.

Every intelligence cycle used to assume a bottleneck: humans reading, humans writing, humans deciding what mattered. AI has removed that bottleneck. It hasn’t removed the need for judgement.

The report’s most useful finding is also its most inconvenient one: the volume problem intelligence teams have spent a decade solving is being replaced by a trust problem that no one has solved yet. And that trust problem doesn’t stay inside the workflow. It travels with the product, all the way to the desk of the person who has to read it, decide something, and act.

Speed without rigor is just noise, faster.

This is where AI in intelligence analysis gets genuinely difficult. Faster production means faster decisions. Faster decisions built on unverified inputs are just confident guesses with better formatting. In open-source intelligence (OSINT), that risk is sharpest. The raw material comes from the surface, deep and dark web, where an unverified source can read exactly like a verified one. The Reading the Room report names this tension directly. It’s the gap between what AI can generate and what a decision-maker can stand behind.

The report puts a number on that gap. A 2023 survey of US policymakers found nearly 60 percent of the intelligence they received arrived in writing, when less than half said that was their preferred format. That’s not a production problem. That’s a consumption problem, and it’s been sitting there for years.

That  consumption gap shows up differently depending on where you sit:

• Analysts feel it as pressure to produce more, faster, without a corresponding increase in time to verify.
• Senior officials and decision-makers feel it as a widening gap between the speed of the briefing and their confidence in it.
• National security and defense leaders feel it as an accountability question: who signs off when the pipeline itself is partly AI-generated?
• Law enforcement and regulatory teams feel it as scale without safeguards, more signal to chase, no more capacity to chase it safely.

Personalization and interactivity aren’t nice-to-haves anymore

Commercial OSINT crossed this line years ago. Customers expect outputs tailored to their specific risk threshold and use case, not a generic feed, and they increasingly expect to interrogate the intelligence directly rather than wait for the next brief. The report’s Scenario 1, a conversational interface for querying intelligence holdings, isn’t speculative. The technical capability already exists in commercial tools. What’s still being worked out is how to do it safely inside a classified environment.

Reading the Room Report lands the key points on how intelligence is consumed

Most AI-and-intelligence writing stops at the warning. The Reading the Room report goes further. It treats how intelligence is consumed, not just how it’s produced,as the actual design problem. None of it replaces human judgement. All of it assumes human judgement stays firmly in the loop.

Why Fivecast is part of this conversation

Fivecast works with agencies and analysts all over the world living this shift every day, and we’ve built our platform around the same conviction the report lands on: AI should give analysts more signal and more time, not more noise to manage alone. That’s the thinking behind Fivecast ONYX, built to cut through digital noise and make OSINT investigations up to 80% faster, with human judgement firmly in the loop.

Cover More. Miss Less.