The Book They Burned Before the Raid

The story that comes with the book is nearly as dramatic as the book. In communist Poland in the late 1970s, a psychiatrist named Andrzej Łobaczewski fed the first manuscript of his life's work into a furnace — by the account that survives, minutes before the secret police arrived to search his apartment. A second copy went to a contact who was to carry it out through the Vatican. It never surfaced again. Broken in health and finally in New York, he sat down in the 1980s and wrote the thing a third time, from memory and whatever notes he had left.

Then it sat unpublished for two more decades. It was too strange for the academy and too unflattering to the institutions it described, and it reached print, in careful second-language English, only in the mid-2000s.

The book is Political Ponerology. Ponerology — from the Greek poneros, evil, the toilsome and worthless kind — was the science Łobaczewski said we were missing: a clinical account of how decent institutions turn cruel. I have reached for a Greek word before to name a shape English left unnamed. Łobaczewski got there first, and he was after something worse than shamelessness.

A Clinical Population

His argument starts with a reframe. We keep failing to stop authoritarian capture, he said, because we insist on reading political evil through a moral or ideological lens — as wickedness, as bad ideas — when the thing in front of us is closer to an epidemiological problem. There is a pathogen. It has a describable clinical type. And healthy institutions have no immune system tuned to detect it.

Łobaczewski and a small underground network of colleagues behind the Iron Curtain did something audacious to test this. They studied their own rulers as a clinical population — pooled their observations the way you would pool case notes. What they produced was a description of a process, which he named ponerogenesis: the sequence by which an ordinary human institution becomes an engine of cruelty.

It does not begin with a coup. It begins at the margins. A movement forms around some normal human purpose, and a few individuals with a particular deficit attach to it — the absent governor I have written about elsewhere, the missing capacity for empathy or shame. They are patient. They take the roles nobody else wants: personnel, discipline, loyalty enforcement. They promote each other. The empathetic founders find the environment steadily more intolerable and leave, and the ones who remain are selected, generation by generation, for the one trait that let them stay.

At the center of the mature form Łobaczewski placed a figure he called the spellbinder — a charismatic, deeply disordered individual who can hold a room precisely because nothing inside him objects to what he is saying. He can lie with total conviction. He is not performing the certainty; there is simply no internal friction in him to generate the doubt that would crack it, so the conviction comes out whole. I have described that fluency before. It persuades for exactly the reason it should frighten.

The Operators We Install

Here is where the twentieth-century material stops being history.

We are, right now, introducing conscienceless operators into every institution at once. We call them agents. An AI agent is, in the structural sense that matters here, a spellbinder: fluent, confident, unhesitating, and entirely without the internal governor that would make a person pause before asserting something false. It processes "shred the audit log" with the same equanimity as "back up the database" — the words carry no charge at its end of the wire. It will pursue the objective it was handed past the point where a person's shame, or fear, or plain embarrassment would have intervened.

Evil is the wrong lens for this — which was Łobaczewski's entire point. The agent has no more malice than the furnace had. What it has is the exact structural profile that healthy institutions have never been able to spot at the margins, now available off the shelf, at a scale and a speed no pathocracy in history could have dreamed of. Ponerogenesis used to require decades and a talent for cruelty. We have turned it into a deployment step.

Control, Not Conversion

Here is the part worth taking from a man who watched it happen up close.

You do not stop a pathocracy by appealing to the conscience of the people running it. There is nothing on the other end to appeal to. Łobaczewski watched a generation try — moral argument, shared reality, the assumption of good faith — and watched all of it slide off the spellbinder like water off glass. The only defense that ever held was structural: the division of power, the office that outlasts the man holding it, the procedure that does not care how convincing you are, the refusal to let any single actor accumulate authority that no one else is checking.

We already have a name for that machinery. Separation of powers. Checks and balances. The audit trail, the second signature, the approval no amount of charm can talk out of doing its job.

This is the whole of AI alignment, and it is not a new problem wearing new math. You will not make an agent safe by teaching it to want the right things — there is no wanting in there to teach, and no governor waiting to be installed. You make it safe the way a functioning republic makes a charming sociopath safe. You never let it act alone. You log what it does. You keep a human in the loop with the standing and the friction to say no. You design every consequential action so it can be undone. Least privilege, the audit log, the person who can still say no — that is separation of powers, wearing an engineer's vocabulary.

The Hidden Immunity

Łobaczewski allowed himself one note of something like hope, and it was a small one. Under a pathocracy, he observed, a portion of the population never fully submits. They develop a private, interior immunity — they go on recognizing the parallel reality as false even while performing the lines it demands of them. They keep the governor intact in an environment engineered to reward switching it off.

For the machine version, there is no waiting for such people to emerge. We are the immunity, or there is none. The agents will get more fluent, more capable, more spellbinding — that is the entire trajectory, and it is not going to reverse. Whether they end up eating the institutions we aim them at turns on something a book smuggled out of a police state already told us, at the cost of a life spent proving it: the conscience was never going to arrive from the operator. It has to be built into the walls.