Two agents each choose to cooperate or defect; whatever the other does, defecting yields a higher individual payoff, so both rationally defect and end up worse off than if both had cooperated. The structure is defined by its payoff ranking (temptation beats mutual cooperation beats mutual defection beats being the lone cooperator), not by any specific story about prisoners.
In a single round defection dominates, but when the same game is repeated with the same partner, cooperation can become rational because future retaliation and reward are at stake. The one-shot verdict and the repeated-game verdict genuinely diverge, which is why real players cooperate more than single-round theory predicts.
In 1950 at RAND, Merrill Flood and Melvin Dresher had economist Armen Alchian and mathematician John Williams play a hundred rounds of the dilemma; the two cooperated far more often than pure self-interest predicted. Told of the result, John Nash observed that rationality in the iterated game can differ from the one-shot case, anticipating the repeated-cooperation insight.
The book reads the Cold War nuclear buildup as a two-player dilemma: each superpower finds building more weapons individually rational regardless of the other's choice, so both arm heavily even though mutual restraint would leave both safer and richer. Seeing the standoff this way reframes the arms race as a structural trap rather than mere aggression.