I know it's not worth anything, but I used to work with an ex-NSA cryptographer (whom I personally knew in meatspace and yes, actually had worked at the NSA) who claimed that they could create purpose-built machines to factor out keys and read the contents of encrypted messages. The target would have to be high-value as this was a costly exercise. He may have been yanking our chain, but for what it's worth, that's what he would claim. He didn't specify which individual ciphers this applied to.

However, there are so many ways to circumvent encryption that it seems that there aren't a lot of cases where a hard decryption of a previously-coded message would really be required (the target can often be compelled or tricked into giving up his keys, a plaintext version may be intercepted at some point, etc.).

Almost certainly not. Most likely no one in the world has that capability, at least as far as I can tell. It would be extremely surprising if any government somehow broke AES without researchers around the world figuring it out within the same general time-frame given some new advancement.

The issue is getting perfect implementation and perfect opsec. With enough time, and willingness an opportunity will find itself.

No, and I doubt that this capability would be revealed for something at the level of San Bernardino, or anything short of "ISIS is going to do a WMD attack against Manhattan".

" assuming good opsec"

Big assumption right thurr. Besides, "good" won't cut it - you'd need "perfect even under adversarial circumstances". Good luck with that.