You're right of course, however we need to make a distinction here. First of all there's the issue of how strong is an encryption algorithm. For example RSA is

You're right of course, however we need to make a distinction here. First of all there's the issue of how strong is an encryption algorithm. For example RSA is based on the problem of factoring large numbers, a problem that's generally considered to be hard as we know of no efficient algorithm for solving it. But we haven't proved that factoring large numbers will remain a hard to solve problem in the future. The NSA could very well have custom hardware for efficiently factoring 1024-bit primes by now and the upcoming quantum computing is a real threat. If they haven't done it by now, 1024-bit keys will become breakable in the future, however 2048-bit keys are another issue entirely and 4096-bit keys will probably stay unbreakable. But, even if breakthroughs in solving the integer factorization problem will be made in the future, as long as P != NP then perfect encryption is possible. In fact, we already know of encryption schemes that are provably unbreakable even with unlimited hardware at disposal, the problem being that they are also hard to implement, so we ended up with making tradeoffs. Second, it's far easier to attack a particular implementation, to bypass the encryption algorithm entirely, e.g. attacks against the key generation system, side-channels, the protocol of the software system we are talking about, etc ... because software always has bugs, as in zero-day exploits that one could make use of. For this reason - if indeed the military is using different encryption algorithms for different security levels, algorithms that aren't used in the wild, then to me that's a pretty bad idea, as far more often than not it's the implementation that's broken, not the algorithm. And in case of inside leaks, the implementation is always easier to get a hold of, compared to the key.