AES provides diffusion at the block level by implementing the ShiftRows and MixColumn sublayers. The former permutes data on a byte level, wheres the latter combines blocks of four bytes.
Can it ever be the case that this level of diffusion is insufficient?
Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. Avoid asking multiple distinct questions at once. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.
That depends on how you define sufficient. AES achieves full diffusion after only two rounds, where diffusion is defined as the ability for every bit in the state to affect every other bit. That does not mean that a mere two rounds can protect against cryptanalysis. The number of rounds necessary to avoid fatal cryptanalysis is something that can only be chosen by studying the cipher and attempting to break it. During the AES standardization process, the cipher Rijndael, which became AES, was given 10 rounds for the 128-bit key size because the best attack against it could break 6 rounds. The extra rounds were considered to be a safety net, or security margin, against future advances in cryptographic attacks.
Is this always going to be sufficient? We don't know. Right now, no attack more efficient than exhaustive search is known that can break every round in the cipher in order to violate its security properties. Some have argued that 10 rounds is not enough, and even suggested that it be raised to 18 or 24 during the AES process, but that never happened and it was standardized as 10, 12, and 14 rounds for each key size. However, as the years have passed, we still do not have any cryptanalysis that breaks every round.
