Too many leaders don't understand AI.

Worse, they don't know what they don't know.

It's time to say the quiet part out loud. We have a comprehension crisis.

I ran a poll recently. I asked people who actually work with AI how well the leaders pushing it understand what they're pushing. Seventy-two percent said "vaguely" or "clueless." Two percent said "very well."

Two percent.

Another survey found similarly large discrepancies. Only 9% of individual contributors believe their managers are prepared to lead on AI. Yet, 80% of the C-suite believe they are. And 71% of CIOs and CTOs (the people actually building AI systems) say their own leadership's expectations for AI are flatly unrealistic.

The people working directly with AI know the decision-makers are faking it.

When you ask confidentially, the people in charge admit it. In one survey, 91% of C-suite executives confessed to pretending they knew more about AI than they did, a higher share than the employees beneath them (79%). The higher you sit, the more you bluff. And yet nine in ten of those same leaders said they personally had all the AI skills they needed, while pointing at their colleagues as the ones who fall short. Everyone is certain that it's the other guy who's unaware.

Ask the board, the supposed grown-ups. Three-quarters of directors rate their AI knowledge at or above their peers'. Yet, fifty-eight percent of them have had no AI training at all. Confidence without competence.

And notice who thinks they are immune to the impact. Seventy-nine percent of executives think AI could replace some of their employees. Yet only 26% think it could touch their own job. Disruption, apparently, is always for other people.

Given this background, it should not surprise us that 95% of enterprise AI pilot projects deliver no measurable impact to the bottom line. Or that 56% of companies say they are getting nothing at all.

Every one of those numbers is self-reported. This isn't critics throwing stones. It's the frightening reality.

If there was ever a time to tell the emperor his wardrobe is lacking, it is now. We are on the plane, but the captain doesn't realize he's lost. We are following a commander who ignores the warnings, even as his advisors describe what's coming.

Two dangerous psychological phenomena are happening at once.

First, we have something worse than cluelessness: being clueless about being clueless. The situation is called Dunning-Kruger, named after the two psychologists who identified it. It is when decision-makers lack the competence to recognize their own incompetence, because the skills you'd need to do it well are the same skills you'd need to know you're doing it badly.

When you aren't aware of how unaware you are, there is no urgency to learn.

Second, at the next level down, we have people who are reluctant to point out the obvious, despite seeing it clearly. Psychologists call it pluralistic ignorance. It's a group of people who each privately doubt, but assume everyone else understands, so no one speaks up. (This concept is taught in psychology 101 using the Emperor's invisible new clothes.)

Their silence gets mistaken for agreement.

AI didn't invent these human tendencies. But it may be the first technology built to reinforce them. It is fluent, agreeable, and endlessly flattering. Asking it a question in a chat and getting a confident answer back feels like competence. Normally the cure for Dunning-Kruger is expertise: get good at something and you start to see your own gaps. AI removes even that. It flatters the expert and the novice alike, so they can reassure themselves that they know how to "use AI."

So you get two kinds of blindness, and they end in the same place. The emperor is sure he gets it. And the peer nods because he figures everyone else gets it, so he doesn't have to: "let IT handle it." One has both hands on a wheel he doesn't know how to steer. The other has taken his hands off the wheel entirely. Either way, no one is driving.

Here's the twist. In the fairy tale, the emperor is only embarrassed. He parades through town, a child says the obvious, everyone laughs, and life goes on.

This version doesn't end at embarrassment.

When no one is driving, people get hurt. Boeing put an automated flight system on the 737 MAX that its own leaders didn't fully understand and didn't scrutinize. Two planes crashed. Three hundred forty-six people died. And a board held personally liable for failing to oversee it: a $237.5 million settlement. The blame did not stop at the machine. It climbed the org chart to the people who were supposed to be watching.

And you cannot blame AI. When Air Canada's chatbot invented a refund policy, the airline argued in court that the chatbot was "a separate legal entity responsible for its own actions." The tribunal called that "a remarkable submission" and held the airline liable anyway. You can try to claim ignorance, but you will be accountable anyway.

Which is, strangely, the good news. Because there's only one way to break the spell: somebody says the true thing out loud. The 72% in my survey (who said their leaders don't get it) have done exactly that.

Leaders don't have to become data scientists. Boards have never understood everything they govern; they oversee a CFO without being accountants. Oversight was never about expertise. It's about the humility to admit you lack it, and the nerve to ask anyway. Govern AI the way you govern money: don't fake the fluency, and don't outsource it to the lowest-level accountant. Make sure someone in the room understands both the AI technology and the underlying business requirements.

Fake-it-till-you-make-it isn't an AI strategy. Neither is "let IT handle it."

So the next time a leader announces a bold, new AI initiative, and the room goes quiet, listen to that small, internal flicker of doubt you're afraid to voice — that flicker is the most valuable thing in the building.

Say it.