Nobody thinks of themselves as the person who would do it.
That’s the first thing worth sitting with. Ask anyone whether they would electrocute a stranger on the instructions of a researcher in a lab coat, and the answer is obvious. Of course not. They’re not that kind of person. The Nazis were that kind of person. The guards were. The people who followed orders. Not me.
Stanley Milgram ran his experiments in the early 1960s specifically because he wanted to test that assumption. He expected it to hold. It didn’t come close.
In his original studies, roughly 65 percent of participants delivered what they believed were severe, potentially lethal electric shocks to another person, simply because a man in a lab coat told them to continue. No threats. No coercion in any dramatic sense. Just a calm voice and an institutional setting and the words: “Please continue. The experiment requires that you continue.”
Sixty-five percent.
That number has been replicated, challenged, refined and debated for sixty years. What it has not been is explained away. The psychology of obedience is one of the most documented and least understood phenomena in the entire field. We know it happens. We still don’t fully know how to stop it.
The Situation Is the Variable Most People Ignore
The instinct, when we hear about atrocity, is to reach for character. There must be something wrong with those people. Something broken, or twisted, or absent. This is not just wrong. It is a defense mechanism.
What Milgram’s work demonstrated, and what Philip Zimbardo’s later research on situational power kept insisting, is that the situation shapes behavior far more reliably than personality does. Move a person into a particular context and give them a particular role, and what they will do becomes dramatically more predictable than who they believe themselves to be would suggest.
This is deeply uncomfortable because it implicates all of us. It means that the question is not whether you are the kind of person who would comply. It means that under the right conditions, with the right framing, with authority present and accountability diffused, most people will.
That’s not a pessimistic claim about human nature. It’s a structural claim about how behavior works. And the difference matters, because pessimism leads to resignation while structural understanding leads somewhere more useful.
The situation doesn’t excuse. But it does explain. And explanation is where any serious response has to begin.
What Authority Actually Does to the Nervous System
There’s a tendency to treat obedience as a purely cognitive process. We hear an instruction from someone in a position of authority, we assess it, and we decide whether to comply. Rational actors making choices.
That’s not really how it goes.
What authority does, at a level that precedes conscious reasoning, is shift what psychologists call the agentic state. Milgram described this himself: participants who obeyed weren’t making active decisions to harm someone. They had slipped into a mode where they experienced themselves as instruments rather than agents. The responsibility, in their perception, had transferred upward. To the researcher. To the institution. To the experiment itself.
This is why obedience under authority rarely feels like a moral choice in the moment. It feels like doing your job. Following procedure. Being professional. The moral weight of the action gets diffused into the role, into the hierarchy, into the system that instructed you. You’re just the person who administered the shock. The researcher carries what it means.
Antonio Damasio’s work on emotion and decision-making is relevant here in a way that doesn’t get cited enough. When the emotional signal that would normally accompany harm gets suppressed or rerouted by institutional context, the warning system that should make you stop simply doesn’t fire loudly enough. Obedience isn’t just a social phenomenon. It has a neurological texture.
Bureaucracy as an Obedience Machine
Milgram’s lab was a controlled setting. The real world operates at scale. And scale requires something more sophisticated than a single researcher with a clipboard. It requires bureaucracy.
Hannah Arendt was covering the Eichmann trial in Jerusalem in 1961 when she coined the phrase “the banality of evil.” What she saw in Eichmann was not a monster. She saw a functionary. A man who processed paperwork, managed logistics, coordinated schedules. A man who, by his own account and largely by the record, was simply doing his job within a system that had normalized what it was doing.
This is the industrialized version of Milgram’s finding. When harm is distributed across enough roles, across enough steps and departments and sign-off procedures, no single person feels fully responsible for the outcome. The person who writes the policy didn’t move anyone. The person who processed the forms didn’t enforce anything. The person who enforced didn’t decide. Everyone followed procedure. Nobody did it.
Modern institutions replicate this structure with remarkable consistency. Not usually toward atrocity, but toward harm that accumulates in smaller increments: financial systems that exploit without individual malice, healthcare bureaucracies where patients fall through procedural gaps, organizations where decisions nobody would individually endorse get collectively produced through a chain of ordinary compliance.
The machinery works the same way whether the outcome is catastrophic or merely damaging. And the psychological protection it offers to each participant is identical: I was just doing what I was told.
What Makes Some People Refuse
Here is the part that gets less attention than it deserves. Not everyone obeyed.
In Milgram’s studies, roughly a third of participants refused, most of them at some point in the process rather than from the start. They did not comply all the way to the maximum shock. They stopped. Pushed back. Told the researcher they were done.
What distinguished them?
Several things showed up consistently. People who disobeyed were more likely to have witnessed someone else disobey first. Social proof works in both directions. The presence of a dissenting peer cut obedience rates dramatically in Milgram’s variations. One person saying “I’m not doing this” made it significantly easier for others to arrive at the same conclusion.
Distance also mattered, in reverse. When participants could see or hear the supposed victim, compliance dropped. Proximity reinstated the emotional signal that bureaucratic distance suppresses. The suffering became real again. The warning system fired.
And there is something less quantifiable that shows up in the accounts of refusers across various contexts: a prior commitment to a value that was experienced as non-negotiable. Not a vague preference. Something that had been thought through, articulated, and attached to identity. When the instruction came, it didn’t just feel wrong. It felt like a clear violation of something they already knew they stood for.
This is not about moral superiority. It is about preparation. The people who resist are often people who have already decided, in calmer moments, what they will and will not do. The decision gets made before the pressure arrives.
The Role of Incremental Escalation
One more thing Milgram built into his design deserves more attention than it typically receives.
The shocks didn’t start at the maximum. They started small. Fifteen volts. Barely perceptible. Then thirty. Then forty-five. Each step was modest relative to the previous one. By the time participants were delivering severe shocks, they had already delivered so many lesser ones that stopping felt inconsistent. They had established a pattern of compliance. Breaking it would mean implicitly acknowledging that all the previous steps had been wrong too.
This is escalation. And it is not a laboratory artifact. It is how compliance is built in the real world, in almost every context where it matters.
Organizations don’t ask new employees to do the most ethically compromising thing on the first day. They ask for something small. Plausible. Defensible. Then something slightly larger. Each step is normalized by the previous one. The increments are calibrated to stay within the range of what feels manageable. By the time someone is doing something they would have refused at the outset, they have a history of compliance that makes refusal feel like a self-indictment.
This is why the question “would you do X?” is so much easier to answer in the abstract than in the middle of a sequence that started with A, B, and C. Moral clarity degrades under incremental pressure. The architecture is designed for it.
What It Takes to Make It Stop
So what actually interrupts obedience? Not the vague aspiration to be a good person. Not general awareness that authority can be abused. Those don’t hold up under situational pressure. What holds up looks more specific.
First: the presence of dissent. A single person willing to say out loud that something is wrong changes the landscape for everyone else. Social proof is one of the most powerful forces in human behavior, and it doesn’t care which direction it runs. Cultures of silence enable obedience. Cultures where dissent is visible and survivable do something different.
Second: structural proximity to consequence. When the people making decisions have to be near the outcomes of those decisions, the emotional signal that authority suppresses tends to reassert itself. This is not sentiment. It’s a design principle. Systems that insulate decision-makers from consequence produce more harm, more reliably, than systems that don’t.
Third: prior commitment. Before the pressure arrives, not during it. Values that are articulated, tested, and attached to something a person actually cares about are more resistant to situational override than values held loosely as abstract preferences. This is one reason that organizations genuinely interested in ethical behavior invest in the kind of conversations that happen before the difficult moment, not just in compliance training that tells people what not to do after the fact.
And fourth, perhaps most importantly: the deliberate disruption of incrementalism. Naming the escalation pattern when it’s happening. Asking not “is this step defensible?” but “where did we start, and how did we get here?” Those are different questions. The second one is considerably harder to answer comfortably.
None of this is a guarantee. Situational pressure is real and it is powerful and it does not disappear because someone has read Milgram. But the gap between knowing how obedience works and knowing nothing about it is not insignificant. It is, in fact, the only ground on which resistance can be built.
The architecture of obedience is not destiny.
It is, however, a very good blueprint for understanding what we’re actually up against.
