My bet is on the core of AES, namely Joan Daemen' and Vincent Rijmen's substitution-permutation network hocus-pocus. A backdoor transformation would suffice to make all other paranoia superfluous.
The smart people like Schneier tend to think that the mathematics of modern encryption is solid. It seems much more likely that a backdoor in an algorithm is detected by academia than a backdoor in some hardware module or in the implementation of some proprietary software.
