Of course, that does not take the elephant in the room into account. By using proper steganography, terrorists could use encryption pretty much unnoticed.

But only if you also keep transmitting junk data over the same mediums. There is plenty of research available for detecting the presence of applied steganography by entropy measurements. [0, 1, lots of others]

If you only apply steganography when you have real[tm] communication needs, those transmissions will stand out. So, from information theoretic standpoint you need to ensure that transmission is undetectable from steady-state with no payload.

Incidentally this is also one part of Pond's threat model. [2]

0: https://ntrl.ntis.gov/NTRL/dashboard/searchResults/titleDeta...

1: https://www.blackhat.com/docs/asia-14/materials/Ortiz/Asia-1...

2: https://pond.imperialviolet.org/

Definitely, but like cryptography, this is very much a cat-and-mouse game. In some data uniform noise is harder to detect than other data. You also have to exploit streams that are large or expensive to analyze. But I think that there are many stochastic processes with near-uniform properties waiting to be exploited (such as e.g. some word order choices in some languages with free word order).

The thing is that with steganography, plausible deniability is much easier than pure cryptography.

Doesn't properly compressed data look just like junk data? You could simply compress some file, add a number of random bits ("corrupted during transfer"), and use just a few of those bits for encrypted communication.