It does, it looks random (hight entropy).

Yes, measured data and unmarked compressed data have this same property, as do actual random data. But is does not look like 9 nines of false positive rate are a concern to those people.

You can set the entropy to any amount you want. You need to consider encryption methods that put in at least a bit of effort to hide themselves. It could select random phrases and pretend to be a spambot.

Or it could look like the output of a Markov chain sentence generator on the topic of religion and post the messages in some well-known forum.

Well, where's the boundary between cryptography that hides itself and stenography? Is there one?

If you include stenography, yes, it's certainly not easily recognizable. I don't think good stenography can be recognized at all, but that's not my area and I've got people contradict me at this (without further info), thus I'm not sure.

There's not much of a boundary, but that wasn't exactly the point I meant to make. You can do something like encode as ASCII 0s and 1s and have low entropy without that hiding anything.