Although I think this would be an edge case of such PL, one way to solve this is to offset this problem to the user. A provided proof for it being safe instead of auto prove it.