Human intervention largely indicates a critical failure.
But the knowledge that it will happen is enough to stop anyone from buying the keys. So it will (likely) never be needed.
I think his point is that once a highly unlikely set of assumptions are accepted as a given then you can correctly make highly unlikely set of conclusions
I actually dont see a flaw with that logic.