"Manafort doesn't know how to play Prisoner's Dilemma!" tweets are probably tongue-in-cheek, but do raise a deep point that when observed behavior is not optimal given your model, the problem usually lies not in the rationality of the actor but in how you modeled them.
Yup. 3rd party sanctions are a well-established way to "beat" the Prisoner's Dilemma. It's an insight derived from the model, not a strike against it.