Overconfidence in the controllability of nuclear weapons creates danger. The passing of the last elite witness of the most dangerous nuclear crisis, that is, the ‘Cuban Missile Crisis’, and the current Trump administration only make this more salient. In this context, this article reviews the scholarly literature about the limits of predictability and controllability of nuclear crises and investigates three failures of learning from them. Given that France displays in particularly acute form some of the sources of overconfidence in the controllability of nuclear crises that can been found in other nuclear armed states, this article offers the first study of the French experience and evolving interpretation of the Cuban missile crisis in comparative perspective, based on untapped primary material. In security studies, this article makes three contributions. First, the publication and interpretation of primary sources is a contribution in itself given the frequent misconceptions about nuclear dynamics due to theory-driven extrapolations. Second, it challenges a widespread assumption of automaticity linking a fear-induced deterrent effect and the presence of nuclear weapons. Third, empirically, this article studies part of a regime of valuation of nuclear weapons. It finally outlines a research agenda to take luck seriously in security studies.