my answer might be late to the party but I'd like to contribute. Depending on whom you ask, you will get different answers, some might even state that there is no difference, but we will look at what we know in order to determine whether such encryption exists, and if it does, if it's any good.

The NSA uses a suite of algorithms called "Suite A" (they actually use more suites, but this is the one apropiate for high-stakes military use), which are mentioned in spec sheets of military cryptoprocessors used by Raytheon. 
Military encryption exists, and that's a fact, not an opinion.
Here's where the subjective opinion starts flowing in. Are military encryption algorithms as safe, or perhaps, better than standard algorithms, such as AES?

I have noticed someone was finger pointing the NSA for violating the Kerkchoff principle, but are they truly?
Kerchoffs principle states that the security of a system should be based on the secrecy of the key, not the secrecy of the algorithm. 
Having this definition, we may look at one cipher that was at one point classified, and that is Skipjack. Skipjack was, for its time, a pretty decent algorithm, back when the NSA was even less experienced with cryptography than it is now. Skipjack, although it was secret, was not designed to have its security based on the secrecy of the system, thus, respecting the principle mentioned before. 

We may now look at Suite A algorithms and ask ourselves if that is still the case. Most cryptographers will normally bash on anyone using secret algorithms, Jean Philippe Aumasson, Niels Ferguson, Bruce Schneier will ALL tell you not to use secret algorithms, but their arguments are based on either Kerkchoffs principle, or on the fact that by having a cipher made public, more people can look at it and analyse it.

Here's where the adversarial mindset comes in : the mentioned cryptographers assume that anyone capable of finding non-generic attacks on a cipher will also publish them, with the purpose of helping the designers of that particular cipher enhance it, but is that truly the case?
If the NSA would publish Suite A tomorrow, and the chinese secret services would start analysing these algorithms, would they publish a paper regarding non-generic attacks in order to help the designers at NSA enhance the ciphers of suite A, and not take advantage?
If the cryptographers at NSA were certain of the fact that AES(we'll come back to AES very soon) can withstand military attacks from the chinese, would they not tell the military to only use AES? 
AES is an amazing algorithm and it's respected, but it's no secret that you get a better security margin if you use non-standard algorithms. Twofish has a better security margin than AES, and so does Serpent, AES is simply outperforming them, which is an important aspect in order to have good data traffic.
In Modern Cryptographic Applications, William Easttom argues that the NSA is probably not publishing secret algorithms in order to avoid having adversaries analyse them. The NSA is also more than capable of analysing the algorithms themselves in order to determine whether they are secure or not. 
There is number of things that, when taken to account, point to one "conclusion" (until suite a is published and analysed, we can't really call it a conclusion) : AES is just the standard, and those who know the story of how it came to be, know there are better algorithms. Also, if the largest employer of cryptographers in the world came to the conclusion that military cryptoprocessors should use both Suite A and Suite B algorithms, classified algorithms are probably safe, given you have enough brains to breed them.