Cognitive Bias in Nuclear Command Which Distortions Bind at the Desk, Which Bind Above It, and Which Are Not Command Problems

"Cognitive bias in nuclear command” names three problems that are routinely collapsed. The first is warning-desk cognition under a short clock: discrimination of machine-mediated reports and revision when a second channel disagrees. The second is principal-and-staff cognition over hours or days: holding competing readings of adversary intent and keeping a reversible act on the menu. The third is public and elite approval of nuclear use, measured in survey experiments. Only the first two are command problems. Named biases are useful when they specify which of four operations fails—discrimination, maintenance of alternatives, two-sided revision, or preservation of a reversible choice—and at which layer of the procedure. They are not useful as a master cause that overrides safeguards. Near-misses show bias present and safeguards holding (NORAD 1979–1980), bias present and a duty officer plus doctrine plus a missing radar holding (Petrov, 1983), and polarized staffs with a principal who refused to close the menu (Executive Committee of the National Security Council [ExComm], 1962). Hardware faults and tightly coupled accidents are not biases. Survey findings on psychic numbing and lesser-evil contrast are not command. The implication for deterrence theory is narrow. Biases are how a change in processing and one-sided revisability appear in ordinary language. Architectures should be judged by whether they keep a second hypothesis live, keep a second channel faster than the commit window, and keep revision open toward accident as well as toward attack.