This from today's NPR story remembering Winston Churchill:
"Historian Andrew Roberts says the impact of Churchill's speeches cannot be underestimated."
Most definitely it is a misstatement (either by the historian or the author of the story), and the last word should have been "overestimated": a way to say that the true dimension of whatever being measured is so great that any estimations of it will risk falling short.
Unless he had meant differently by deliberately choosing "underestimated," for instance, that one should be mindful of how impactful Churchill's speeches are.
Or, heaven forbid, he meant that the impact of the speeches are minimal.
The mistake you cite is a pet peeve I share, Robert. But I offer an alternative interpretation, one that triggers another pet peeve of mine: Maybe he meant the impact of Churchill's speeches must not be underestimated. Americans—maybe other Anglophones, too, but Americans for sure—seem to use "can't" when they mean "mustn't" more often than not. It bugs me.