If one bounds the risk from HIV from an undetectable partner to an unmeasurably small epsilon, dwarfed by other risks by many orders of magnitude--- the edges where the assertion are false don't matter.