Earlier this year, one statistic was hard to avoid. Nearly half of Gen Z employees, it said, were sabotaging their own company's AI rollout. Fortune ran it. It moved fast across LinkedIn. For a week or two it turned up in slide decks and board conversations as shorthand for a problem people already half-believed: the younger ones won't get on board.

It is a good headline. It confirms a fear, it carries a number, and it points the finger somewhere comfortable. So I did the slow thing and read the 37-page report behind it.

The number only holds up if you accept how the study defined its central word.

What Counted as Sabotage

Here is what a person had to do to be counted as sabotaging their company's AI:

Do any one of those, once, and you were in the count. By that measure, the colleague who prefers ChatGPT to the clunky in-house tool is a saboteur. So is the one who missed a lunch-and-learn. The word carrying the whole headline turns out to describe behaviour most of us would call a normal Tuesday.

It also helps to know who paid for the research. The survey was commissioned by a company that sells enterprise AI software, and run by a firm whose business is producing this kind of sponsored study for corporate clients. That does not make the data worthless. It does mean the framing was never neutral, and the definition of sabotage was doing a job.

What the Data Actually Showed

The most telling part sits in the report's own figures. When people were asked why they reached for an unapproved tool, the most common answer was not resistance, or fear, or a wish to watch the technology fail. It was, more or less, that they would use whatever it took to get their work done.

That is not sabotage. That is someone trying to do their job around a system that slows them down. Read properly, a good part of the "sabotage" in the headline is the opposite of resistance. It is people adopting AI so readily that they have outrun the tools their company approved.

"The behaviour the survey called sabotage was, in many cases, people trying to do good work despite the tools they had been given."

None of this means resistance to AI is imaginary. It is real, and in plenty of organisations it matters. But the interesting question was never whether some people drag their feet. It is why. And the honest answer sits higher up than a survey about Gen Z wants to look.

The One Line the Report Got Right

For all that, the report reaches a conclusion worth keeping. AI adoption at scale, it says, is no longer a technology problem. It is a leadership one. That is correct, and it is what the consultancies keep finding from other directions - BCG putting most of the value of AI in people and process rather than the tools, MIT reporting that the large majority of corporate AI pilots produce nothing measurable, and rarely because the models were poor.

So far, so sensible. Then look at what the report tells leaders to do about it.

Set clearer KPIs. Tighten governance. Approve a better platform. Run more training. It is a reasonable list, and every item on it shares one quality: a leadership team can mandate all of it without anyone having to say anything uncomfortable. You can issue a KPI from your desk. You can sign off a governance policy in a meeting. You can approve a platform and instruct a training programme without once turning to the person beside you and asking the harder question.

"Every fix on the list can be ordered from the top. Not one of them asks the leadership team to look at itself."

The Fix That Isn't on the List

The thing the report never mentions is the one thing that cannot be delegated. Before a leadership team can lead a change of this size, it has to have a conversation most teams skip - with each other, not with the workforce.

That conversation is not about agreeing the message. It is about making the tensions underneath it visible enough to lead from. What concerns have gone unsaid. What are we not naming because it would be politically awkward. Whose expertise stops being scarce if this works. And the question nobody puts on a slide: who loses standing if it succeeds.

These are not technology questions, and they are not comfortable ones. A vendor selling adoption software is never going to put them in a report, because they are not a feature you can buy and they do not flatter the person reading. They are also, in my experience, where the real drag starts - not on the front line, and not in the tool, but in a room at the top where everyone agreed out loud and a few people disagreed in private. The resistance lower down is usually the echo.

Whether AI arrives in an organisation as a threat or an opportunity is not decided by the comms pack. It is set by the conditions the leadership team creates. And those conditions get set, or missed, in the conversation the headline would rather we didn't have.