Here's an issue I ran into on this MathOverflow thread: https://mathoverflow.net/questions/18787/montagues-reflection-principle-and-compactness-theorem/18804#18804
Likewise, see Joel David Hamkins's answer here: https://mathoverflow.net/questions/15685/is-it-necessary-that-model-of-theory-is-a-set/15713#15713
A final subtle point, which I hesitate to mention because it can be confusing even to experts, is that it is a theorem that every model M of ZFC has an object m inside it which M believes to be a first order structure in the language of set theory, and if we look at m from outside M, and view it as a structure of its own, then m is a model of full ZFC.
In short: we can prove using Levy Reflection that for any model M of ZFC, some element of M satisfies ZFC (although the model may not believe that the structure does). But this seems like a problem for the axiom of foundation: can't we construct a downward chain of membership composed of the first model, the model which is an element of that model, and so on?
The only idea I've come up with is that somehow the descending chain of these submodels is not an element of the starting model M. But it's not clear to me why we couldn't guarantee that it is, by taking some sufficiently expansive initial model like the Von Neumann Universe. Is there a principle reason that the downward chain would not be a member of the universe? Or is there another way of resolving the issue that I've missed?