In a game of Russian roulette you might have a 100% success rate after 5 shots. But that doesn’t mean the 6th shot won’t be fatal.
In the MAD game, both the goal and the incentives are rationally aligned to avoid pulling the trigger at all. If MAD fails, it will happen because a mistake occurred. If the "Just let the UN take care of it" approach fails, it will fail for the same reason it has failed so many times before: because someone rationally calculated that the consequences were worth it. These scenarios don't really compare to a simple game of Russian roulette.
Logically speaking, it seems best to keep doing what we've been doing, but work as hard as we can to reduce the odds of a mistake.
That is the flaw in your logic. It is assuming a favorable distribution when the distribution is unknown.
It would be great if we had more options, no argument there.