Years ago, I was running a human source. Every time we met, he gave me accurate information: the names were correct, the meetings really happened, the documents were genuine.
My confidence grew with each meeting. Reasonable enough. Except there was another possibility.
My source had access to ten pieces of information. He gave me three. All three were true.
But why those three?
Maybe he judged the rest irrelevant. Maybe he was protecting someone. Maybe, in his position, he never saw certain things. Maybe someone above him controlled what he was allowed to see. Or maybe I had trained my own source to confirm the picture of reality I already held.
Every piece of information I received could be perfectly accurate. And I could still be wrong.
The same thing can happen to you.
Your team tells you a competitor is in trouble.
Revenue is down 12%. True.
Their CFO resigned. True.
They closed two offices. Also true.
You conclude: the company is weakening.
What you don't see is that they deliberately exited a low-margin business, are hiring aggressively somewhere you aren't looking, and are six weeks away from launching a product directly against you.
Nothing you were told was false per se.
But the picture you built was.
You don't make decisions from reality; you make them from what your environment has made observable about reality.
This is a cognitive security problem.
Before information even reaches you, something has already determined what becomes visible, what counts as relevant, what gets measured, what gets said and what... disappears.
Before information even reaches you, something has determined what becomes visible, what counts as relevant, what gets measured, what gets said and what... disappears.
In what follows I want to show three things:
- First, that the system selects before you evaluate.
- Then, how to detect that it is editing reality for you.
- Finally, why the most obvious way of fixing the problem can actually make it worse.
The system selects before you evaluate
A CEO receives positive reports from his executive team. The numbers are accurate, nobody is falsifying presentations, the executives are competent. Everything looks fine!
Now look at the environment producing those reports.
What happened the last time somebody brought him bad news? Who gets promoted? What determines bonuses? Which problems are expected to be solved before reaching him? How many layers separate him from customers? What does everyone already know he likes to hear?
Suddenly, the reliability of each individual report tells us much less about the reliability of his overall picture.
The same mechanism exists outside companies.
A newsfeed can contain nothing but factually accurate stories and still produce a distorted perception of the world if certain events are disproportionately selected and repeated.
Five advisers can independently tell you the same thing while ultimately relying on the same source.
Ten analysts can reach similar conclusions because they use the same datasets, assumptions or AI systems.
Consensus can be real. It can also be an artifact of the environment producing it.
How you know it's happening
There are three main signals.
The first is excessive convergence.
When everyone starts agreeing, confidence rises. But agreement is only strong evidence when the observations producing it are genuinely independent.
Five sources repeating one underlying source are not five confirmations. Neither are five executives looking at the same dashboard. When consensus appears, ask how the consensus was produced.
The second signal is that bad news generally arrives late.
Bad news exist in every organization. If you repeatedly discover problems only when they have become impossible to ignore... you've got a problem.
Somebody probably knew earlier. Information became harder to transmit, for some reason: perhaps people wanted more certainty before escalating it? Perhaps managers tried to solve it themselves? Perhaps nobody wanted to be associated with the problem? Perhaps the problem was already brought up but every hierarchical layer softened the message until it reached you?
The information arrived too late to be useful.
Which leads to a third signal: you keep being surprised by things other people already knew.
When you investigate, you discover that someone didn't miss it. A salesperson knew. A junior employee had seen it. The organization possessed the information.
The decision-maker didn't. The problem comes from the information architecture you built.
To these three signals, add a paradox of power:
The more decision-making power you acquire, the less reality you encounter directly.
At first a founder talks to customers.
Then reality becomes a chain: customer → sales → manager → VP → report → dashboard → founder.
All of your intermediaries are very competent. They're not, individually, the problem. The problem is compression. Every layer must decide what matters enough to preserve and what can safely be discarded.
The same thing happens outside companies, in geopolitics and the way international events are related to us:
An event occurs, the journalist from a wire agencies describes it. Another journalist copies it, and edits it. The editor above transforms it. Then the platform distributes it, the algorithm selects it, the AI summarizes it. You consume... a summary. You are now several transformations away from the original observation.
That is not inherently bad. Large organizations could not function otherwise. But it creates a vulnerability leaders rarely audit: the growing distance between the world in which decisions have consequences and the representation of that world on which decisions are based.
This is where the problem becomes dangerous.
You can have excellent advisers, reports and data. No deception anywhere in the system and still... make a disastrous decision.
The failure can happen entirely upstream, in what your information environment selected for you before you ever began thinking. And unfortunately, the most obvious solution can make the problem worse.