How Leaders Accidentally Destroy Psychological Safety

Psychological safety has won the argument. It appears in values statements, town halls and leadership programmes, and it is hard now to find anyone willing to speak against it.

Agreeing that it matters turns out to be the easy part. The harder question is when a team loses it — because that almost never happens on a day anybody would mark.

Safety is rarely destroyed by someone setting out to destroy it. It gets withdrawn in ordinary moments, by capable people under pressure, and the person withdrawing it is usually the last to know — because they are not the one deciding whether to speak next week.

There is a second difficulty underneath the first. The instrument most often used to watch for this is the one least likely to show it.

The Question Worth Writing Down

In Chapter Six of the book, David's mentor tells him to measure the trust. “Once trust becomes a number,” Rajiv says, “no executive can hide from it, and no team can pretend it's fine.”

What interests me more is what David writes in his notebook the next morning.

How do we measure psychological safety without killing it?

Most measurement programmes never ask that. They go straight to instrument design and treat measuring as neutral observation. It is not neutral. Asking a team to rate how safe they feel is itself a test of how safe they feel.

What he builds is deliberately small. Weekly, anonymous, three statements rated one to five.

“I can admit mistakes without fear.”

“My ideas are challenged constructively.”

“New ideas are welcomed.”

The first week comes back honestly bad — low threes, high twos. He puts it in front of the team without defending it. “This is where we are. Not where we stay.”

What Actually Sets Culture

Edgar Schein spent a career on how culture gets set, and his answer is a list of six things leaders do. What they pay attention to, measure and control. How they react to crises. How they allocate resources. What they deliberately model. How they allocate rewards and status. Who they recruit, promote and push out.

Notice what is missing. Values statements, town halls and culture decks are not on that list — Schein files those as secondary, and they only work when they agree with the six. Against a contradiction, they lose.

That gives a precise answer to where safety goes bad. It goes bad in the gap between what a leader measures and what that leader rewards. You can run the dashboard faithfully every week and teach the opposite lesson on promotion day.

Here are five moments where that gap tends to open.

1. The First Punished Disclosure

Somebody raises a real problem at a bad time. The reaction does not have to be a reprimand. A tightened jaw. A question about why this is only coming up now. A follow-up meeting they were not invited to.

Twelve people are watching, and they update further than the event warrants. The person who spoke paid for that lesson. Everyone else got it free, which is why one incident travels faster than a year of encouragement.

The tell afterwards is not that people start disagreeing with you. It is that problems start arriving late — and late is when they are expensive.

2. The Promotion That Contradicts the Speech

Promote the person who hit the date by quietly sitting on a problem, and you have answered the safety question for everyone who watched.

One promotion can overwrite a year of statements. A speech is an opinion about what the organisation values. A promotion is evidence.

People are not being cynical when they read it that way. They are being empirical.

3. Favouritism, Which Nobody Experiences as Favouritism

Schein's fifth mechanism is rewards and status. Rewards get audited. Status does not, and that is where most of the damage happens.

Research going back to 1975 on leader–member exchange found that leaders do not form one relationship with a team. They form a different one with each person. An inner group gets more trust, more latitude, more access. Everyone else gets a more formal version of the same manager.

The useful part is the explanation. It is not put down to malice, or even to preference, but to a resource constraint — a leader has finite time and attention, and spends it where it has paid off before. That is why this is close to universal, and almost invisible from the inside.

No leader experiences this as favouritism. You experience it as judgement. She has been right four times running, so you weight her view more heavily. Each instance is defensible. Most are correct.

The team experiences a queue, and knows whether it is in it.

This does more damage than the other triggers and is harder to undo. The first one teaches people that bad news is expensive — survivable, because you can still choose what to raise. Favouritism teaches people that their contribution is discounted whatever it contains. There is no strategy available, because the thing being discounted is not the idea. It is them.

That is where capable people stop contributing and start complying.

It costs the leader too. An inner circle is a filter. You hear a great deal, you feel well informed, and most of it arrives from four people who share your assumptions — which is largely why you trust them.

And It Will Not Show Up in Your Numbers

Favouritism does not lower the average. It splits the distribution.

The inner group rates four and five, because for them the statements are simply true. Everyone else rates two. The mean lands near three and a half, reads as fine with room to improve, and gets a slide saying progress is steady.

A team sitting flat at three point four and a team split between fives and twos produce the same number and have nothing in common. One needs encouragement. The other has two populations in it, and only one of them is in the room in any meaningful sense.

So read the spread before the mean. If your instrument reports only an average, it is hiding the trigger most likely to be running.

4. The Quarter With No Slack

Safety is cheap when there is slack. When a deadline tightens or headcount falls, the cost of raising a problem rises for everyone — including the leader, who now has nowhere to put it.

Nobody announces a change. The meeting simply gets shorter, the agenda fuller, and the pause where someone might have spoken quietly disappears. This is why scores fall in quarters when nothing cultural happened at all. Something structural happened, and the culture was the first thing spent.

5. Inconsistency

A leader who is open on a calm Tuesday and sharp under deadline has not created a culture with some bad days in it. They have created a condition where everyone must first work out which version of you is in the room.

That guess has a cost, and the safe answer when you are unsure is silence. Consistent and slightly cool beats warm and unpredictable, which is not what most of us want to hear.

Seniority Makes All Five Worse

There is a version of this argument that says power corrupts. I have argued elsewhere, in Power Doesn't Corrupt. It Reveals., that it does something different — it removes the friction that was holding behaviour in place.

Applied here, that has two consequences, and neither requires anyone to become a worse person.

A mid-level manager's flash of impatience is absorbed by the room. The same flash three levels up is read as policy, repeated in meetings you are not in, and acted on by people who were never there.

And the more senior you are, the fewer people will tell you the room went quiet when you spoke. The signal that you have damaged something is the one your position suppresses.

Why the Dashboard Hides It

Which brings us back to Rajiv's instruction, and the trouble with it.

He is right that a number is harder to hide from. He is also describing precisely why it will be managed. A score nobody can hide from is a score somebody has a reason to move, and moving a score is far easier than moving the thing it describes.

Donald Campbell made the general version of this point in 1979: the more a quantitative social indicator is used to make decisions, the more it is corrupted — and the more it corrupts the process it was meant to monitor. A weekly trust score is soft, self-reported, sensitive to tone, and tied to the manager being measured.

And it needs nobody dishonest. A manager under pressure on a trust score does not falsify it. They mention it more often. They ask the team, warmly, to remember the good weeks. They describe it as something everyone owns together. Each of those is a reasonable thing to do. Together they lift a number while nothing underneath it moves.

So the instrument conceals favouritism inside its average, and degrades the moment anyone is judged on it. That is a narrow path to walk.

What Protected David's Version

His survives because of one decision that is easy to read past. He gave the result back to the people who generated it. The bad first week went to his team, in a room, as a shared starting position. It did not travel up the chain as evidence of how his unit was performing.

The direction the result travels is the whole safety mechanism. A number moving down is a mirror — people can argue with it, recognise themselves in it, watch it move. A number moving up into a review pack is a verdict, and people answer verdicts strategically. That is rational, and it kills the instrument inside two cycles.

The Version I Know From Manufacturing

The same failure is familiar on a shop floor, with no survey involved.

Take any quality indicator — first-time-right, defect escapes, rework hours. As a diagnostic it is genuinely useful. Then it becomes a target, and something quiet happens. Nobody fakes a number. The definition moves instead. A borderline item that used to be logged as a defect becomes an adjustment. A twenty-minute rework becomes in-line correction. A category gets split so the ugly half sits somewhere else.

Every reclassification is defensible on its own. Together they produce a metric that improves while the product does not.

About the Numbers in the Book

Chapter Six goes on to describe a pilot with substantial improvements — safety scores, resolution times, turnover, savings. I am not quoting those figures, and I want to be open about why. They are narrative. They belong to the story rather than to a measured study, and repeating them as outcomes would put numbers into circulation that nobody can trace back to anything. That is the habit this article is arguing against.

The Line That Is Not On Any Dashboard

In the book, a junior engineer named Mai says something in a retrospective that keeps David awake. “For the first time, I feel seen.”

That cannot be gamed into existence. No amount of managing the metric produces it. And it arrived in a comment box, not in a score.

The score tells you whether to look. The comments tell you what you are looking at.

What I Would Actually Do

Three questions, not thirty. The fourth is always tempting and always costs you completions.

Return the result to the team before it goes anywhere else, and say out loud what you make of it. The first bad week is the only chance you get to prove that honest answers are survivable.

Keep it out of performance reviews permanently — and say so, because the suspicion that it might end up there is enough on its own to corrupt it.

Read the spread, not the average.

Look at your last four promotions and ask what someone watching would conclude about raising problems. That answer is your culture, whatever the dashboard says.

Ask who spoke in your last five meetings. If the same four names carry it, you do not have a quiet team. You have an inner circle, and everyone outside it already knows.

And watch for the one signal that means it is already gone: scores that go uniformly high and stop moving. A healthy team has a bad week when something bad happens. A flat excellent line means people have worked out what the number is for, and have started being kind to it.