That is true, but it should not be assumed that a spec. is a thing of fixed validity. A formal spec. has a reasonable likelihood of being less ambiguous, incomplete or self-contradictory than an informal one, or whatever substitutes for a spec. in an agile approach.
That only tells me a spec and its formal verification are necessary but not sufficient assets. Not "we can't prove everything so we'll keep our millions riding on half-assed piles of garbage.