The Perfect Scapegoat

Picture the scene.

A scrum team. A sensitive IT project. Late Tuesday night, a deployment goes out. By Wednesday morning, it's clear something went badly wrong. The kind of wrong that has the CIO's face turning a particular shade of red. The kind that fills a meeting room with people who are suddenly very interested in their shoes.

Then someone speaks.

"It's not us. It's the AI."

And the room — almost imperceptibly — exhales.

— ◈ —

That exhale is what I want to talk about.

Not the deployment failure. Not the technical post-mortem. Not even the question of who, in any legal or contractual sense, is responsible. I want to talk about that exhale. Because in it, something important just happened — something that has very little to do with technology and everything to do with how human beings under pressure protect themselves from shame.

The oldest move in the room

Every organisation has had its version of this person. The one who, when things go wrong, becomes the target. Not necessarily because they caused the problem — though sometimes they did — but because the group needs somewhere to put its discomfort. Someone has to carry the weight of what went wrong so everyone else can get back to breathing normally.

When that person eventually leaves — resigns, gets managed out, transfers to another department — there's a brief, collective sense of relief. And then, quietly, someone else begins to fill the same role. The dynamic doesn't leave with the person. It stays in the room, waiting.

This is not a character flaw. It is a group behaviour under anxiety. When the primary task feels threatened — when the project is burning, when the CIO is red-faced, when careers feel like they might be on the line — groups look for a place to put their fear. Blame is a form of relief. It narrows something overwhelming and diffuse into something small and locatable.

The question has always been: who is available to carry it?

Enter the ideal candidate

The AI won't resign.

It won't get emotional in the meeting. It won't threaten to escalate. It won't go to HR. It won't remember being blamed last quarter and carry a quiet resentment into the next sprint. It won't write a Medium post about toxic workplace culture on its way out the door.

From the perspective of a group looking for somewhere to put its anxiety, the AI is almost too good to be true. It is infinitely available. Permanently blameable. And — crucially — incapable of disrupting the projection.

This is not a prediction about some dystopian future. It is already happening in rooms like the one I described. The attribution "the AI got it wrong" is becoming a reflex — not because it is always false, but because it is so extraordinarily useful. It discharges the tension in the room. It protects the humans whose decisions shaped the outcome. It closes the meeting faster. And it carries no social cost whatsoever.

We are watching, in real time, the emergence of the corporate world's perfect scapegoat.

The culture that's building the trap

Here is where it gets uncomfortable.

Most corporate cultures — and I am not being cynical, I am being descriptive — are structured around blame as an organising principle. Not explicitly. Nobody writes "we assign blame here" in their values documentation. But the underlying architecture is punitive. Post-mortems are exercises in locating fault. Performance reviews punish failure. Leaders who surface problems are managed as problems. The unofficial rule, understood by everyone, is: don't be the one holding the grenade when it goes off.

In that environment, psychological safety — the genuine kind, where people can say "I made a mistake" or "I don't understand this" without career consequence — is not just rare. It is structurally discouraged. The culture selects against it.

Now into that culture, we are introducing AI systems that make consequential decisions. Hiring screens. Credit assessments. Deployment pipelines. Diagnostic tools. These systems don't operate in a vacuum. They are trained on human choices, shaped by human assumptions, deployed under human governance — or the absence of it.

The crystal ball here is not complicated. When those systems produce bad outcomes — and they will, because all systems do — the punitive culture will not turn inward and ask hard questions about the human decisions that shaped the AI. It will do what it has always done. It will find somewhere to put the weight. And the AI, sitting there wordlessly, incapable of self-defence, will be irresistible.

The culture is not accidentally building the blame vector. It is building it with remarkable precision.

What MIT Sloan got right — and what it missed

A recent article in MIT Sloan Management Review used the 2018 Uber autonomous vehicle fatality in Tempe, Arizona as its opening frame. A pedestrian died. The world asked: who is responsible? The safety driver? The engineers? Uber leadership? The regulators?

The authors' answer — which I think is genuinely useful — is that in the age of AI, responsibility needs to be understood as distributed across networks of human and machine action. Old-style blame, the search for a single culprit, doesn't map onto these systems. What's needed instead, they argue, is what they call narrative responsibility: a collaborative, learning-oriented process of constructing the full story of what happened — not to punish, but to understand.

That framework is right as far as it goes. The problem is that it assumes organisations want to construct that story.

Most don't. Not because their people are malicious, but because the culture they've built makes full story-telling dangerous. Narrative responsibility — the honest, trust-based, non-punitive unpacking of a complex failure — requires psychological safety as a precondition. It requires people to say "here is where my decision contributed to this outcome" without that sentence being used against them later. It requires leaders to model accountability rather than deflection.

In a punitive culture, asking people to construct an honest shared narrative is like asking them to hand over evidence in their own prosecution. They won't do it. They'll do something else instead. They'll point at the AI.

What the exhale is actually telling you

Back to that room. The deployment that went wrong. The CIO's red face. The shoes. And then someone says "it's not us, it's the AI" — and the room exhales.

That exhale is diagnostic information.

It is telling you that the group experienced the failure as a threat severe enough to require a blame vector. It is telling you that nobody in the room felt safe enough to say "here is where I made a call that contributed to this." It is telling you that the AI just became the group's primary way of managing its own anxiety — which means the actual causes of the failure, the human decisions and assumptions woven into the system, are about to go unexamined.

And an unexamined failure, in a complex system, is a failure that is almost certain to happen again.

The leadership question

The MIT Sloan authors are right that we need better narratives. But narratives don't build themselves. They require a particular kind of leadership — one that is willing to walk into the room after the deployment disaster and do something almost nobody in corporate culture actually does: model accountability without deflection.

Not "the AI failed us." Not "the team let me down." Not even the passive corporate classic, "mistakes were made."

Something more like: "Here is the decision I made that contributed to this. Here is what I didn't know, or didn't ask, or didn't pressure-test. I want to understand what everyone else was navigating, because I think the full story is more complicated than any single point of failure — and I want us to learn it together, without it being used to punish anyone in this room."

That is not cloud-cuckoo-land idealism. It is the only approach that actually produces organisational learning. It is also, in most corporate cultures, an act of considerable courage — because the culture will resist it. The culture is set up to punish exactly that kind of honesty.

But the alternative — and this is what's coming, at scale, across every sector where AI is being deployed — is organisations that have found the perfect mechanism for never learning from their failures. A blame vector that never fights back. A scapegoat that always shows up for the meeting.

The AI didn't fail you.

The room did.

And the room will keep failing, in exactly the same ways, until someone is willing to say so.