In most organisations there’s someone who understands why things fail. Where the real dependencies sit, what the customers actually need, why the last three initiatives didn’t deliver. And in a lot of those organisations, that person has stopped saying so.
Not because they’ve left, though sometimes they have. Because they’ve learned what’s safe to say and settled into saying that instead.
Why this matters for AI specifically
AI depends on people being willing to challenge its output. A model produces an answer. Someone has to be willing to look at it and say that’s wrong, or that doesn’t smell right, or we shouldn’t act on that. In a team with psychological safety, that happens. In a team without it, the wrong output goes through and nobody says anything.
That’s not a soft concern. It’s the difference between an AI system that gets corrected and one that quietly compounds its own mistakes because the people who could see the problem have decided it isn’t their job to raise it.
How trust breaks around AI
AI transformation often gets handled like any other big change. Decisions made by a small group, implementation pushed downward. The difference is that AI arrives with a broadcast. Announcements, enthusiasm, external comms about innovation and ambition. And it lands on people who are quietly worried about their jobs.
The gap between that upbeat signal and the absence of any honest conversation about what it means for the people receiving it is where trust goes. Questions about roles and headcount get deflected or ignored. People notice. They draw their own conclusions, and they stop volunteering the knowledge that would make the whole thing work.
The signs are quiet
This rarely looks dramatic. The organisation still functions. Metrics get hit, tools get used, retrospectives happen. But the honesty in the room has degraded. People have worked out what’s safe to say and are saying that. The voices that go quiet first are usually the ones with the most to offer, the people who understand why things fail and have stopped bothering to explain.
Some of them leave. The gap between the official version of how the AI programme is going and the real version, the one that exists in side conversations, grows steadily, and nobody with the authority to close it knows how wide it’s got.
What I look for
When I assess this domain, I’m listening for a specific thing: what actually happens when someone raises a concern about an AI output. Not what the policy says happens. What happens. Whether the people most affected by an AI initiative were involved in designing it, or handed it. Whether there’s an honest, shared understanding of what AI means for roles, or a careful silence around the subject.
The careful silence is the tell. It means the narrative has left leadership’s hands, and the people who could make AI work have decided to watch it fail instead.