Fuzzing is a statistical technique that isn't ever going to give you a reassurance that a problem doesn't exist. It's great at giving you counterexamples, so fuzzing is great for discovering vulnerabilities, but unless you're fuzzing your program's entire state-space (which is absolutely impossible for even relatively small programs) then you're not comparing like with like.
The comparison should be formal proof of correctness vs. fuzzing using the formal specification as a source of properties to be tested.