Post-hoc rationalisation
Reasoning produced after a judgment has already been reached, in order to justify it — rather than the process that actually produced the judgment. On this view the conclusion arrives first, by intuition or some other hidden route, and the spoken reasons follow as a kind of press secretary explaining a decision made elsewhere. The same gap shows up in machines, when a system gives a plausible account of an answer that omits what truly drove it.
Reasoning generated after a judgment has been reached, to justify it, rather than the process that produced the judgment. The formal statement is Haidt's social intuitionist model (2001), in which moral judgments arrive by intuition and conscious reasoning is recruited afterward — reason as the press secretary, not the decision-maker.
Post-hoc rationalisation is the narrower cousin of confabulation, specific to reasoning that defends a conclusion already reached. Haidt builds the case partly on moral dumbfounding — people holding to a judgment while admitting they can supply no reason for it — which shows the report coming loose from whatever produced the verdict. His title metaphor puts verbal reasoning as the tail and intuition as the dog.
In the machine half of the topic the same structure appears as a failure of explanation faithfulness: Turpin's language model writes a justification for a biased answer that never mentions the bias. The parallel is why the human-confabulation sources sit in the same topic as the machine-interpretability ones — a stated reason that postdates and misrepresents its cause is a problem common to both.